{"vintage":"Snapshot built 2026-10-09, publications 2025-10-01 to 2026-05-22","scope":"CPC G06N, CPC-verified via Google Patents detail pages","count":883,"patents":[{"publication_number":"KR20260074229A","title":"Lane-Level Traffic Hazard Verification and Multi-Channel Dissemination System and Method Based on a Traffic Hazard Verification Gate","abstract":"본 발명은 복수의 단말·앱·차량·인프라 장치로부터 수신되는 사고, 고장, 정차, 역주행, 낙하물 등 교통위험 후보 데이터를 검증한 후 차선 단위로 전파하는 교통위험 검증·전파 시스템, 허브 서버 및 방법에 관한 것이다. 허브 서버는 온디바이스 AI 탐지 결과, 제보·신고 전후의 이동데이터, 위치·속도·진행방향·도로명·링크 식별자·차선 후보 및 등록 인프라 기준점 정보를 이용하여 위험 후보 데이터의 주행 정합성을 판단하고, 검증완료 조건 충족 전까지 외부 송출을 제한한다. 이후 관제요원 입력, AI 자동 검증 결과 또는 대체 검증 기준에 따라 표준 교통위험 데이터셋을 생성·송출함으로써 허위·중복·좌표 오차 제보에 따른 오정보 전파를 줄이고, 검증 출처, 가공 제한 정보 및 감사로그를 통해 교통위험 정보의 일관성과 추적성을 높일 수 있다. The present invention relates to a traffic risk verification and propagation system, a hub server, and a method for verifying candidate traffic risk data, such as accidents, breakdowns, stops, driving in the wrong direction, and falling objects, received from a plurality of terminals, apps, vehicles, and infrastructure devices, and then propagating the data on a lane-by-lane basis. The hub server determines the driving consistency of risk candidate data using on-device AI detection results, movement data before and after reports, location, speed, direction of travel, road name, link identifier, lane candidates, and registered infrastructure reference point information, and restricts external transmission until the verification completion conditions are met. Subsequently, by generating and transmitting a standard traffic risk dataset based on control center input, AI automatic verification results, or alternative verification criteria, the dissemination of misinformation caused by false, duplicate, or coordinate error reports can be reduced, and the consistency and traceability of traffic risk information can be enhanced through verification sources, information with processing restrictions, and audit logs.","assignee":"이지민","inventors":["이지민","이주상"],"publication_date":"2026-05-22","filing_date":"2026-05-05","priority_date":"","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/40","G","G01","G01P","G01P15/00","G","G01","G01S","G01S19/00","G01S19/01","G01S19/13","G01S19/14","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26","G","G08","G08B","G08B25/00","H","H04","H04L","H04L67/00","H04L67/01","H04L67/12","H","H04","H04N","H04N7/00","H04N7/18","H","H04","H04W","H04W4/00","H04W4/30","H04W4/40"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260074229A/en"},{"publication_number":"KR20260073329A","title":"Electronic device and controlling method of electronic device","abstract":"사용자 인터페이스를 통해 사용자 음성에 대응되는 텍스트 정보를 제공할 수 있는 전자 장치 및 이의 제어 방법이 제공된다. 구체적으로, 본 개시에 따른 전자 장치는 적어도 하나의 오브젝트를 포함하는 이미지가 획득되면 이미지를 분석하여 이미지에 포함된 적어도 하나의 오브젝트를 식별하고, 사용자 음성이 수신되면 사용자 음성에 대한 음성 인식을 수행하여 사용자 음성에 대응되는 텍스트 정보를 획득하며, 이미지에 포함된 적어도 하나의 오브젝트 중 사용자 음성에 대응되는 오브젝트를 식별하고, 디스플레이 상의 영역 중 사용자 음성에 대응되는 것으로 식별된 오브젝트에 대응되는 영역 상에 텍스트 정보를 포함하는 메모 UI (User Interface)를 표시한다. An electronic device capable of providing text information corresponding to a user's voice through a user interface and a method for controlling the same are provided. Specifically, the electronic device according to the present disclosure analyzes an image containing at least one object when the image containing at least one object is acquired to identify at least one object included in the image, performs voice recognition for the user's voice when the user's voice is received to acquire text information corresponding to the user's voice, identifies an object among at least one object included in the image that corresponds to the user's voice, and displays a memo UI (User Interface) containing text information on an area on a display corresponding to the object identified as corresponding to the user's voice.","assignee":"삼성전자주식회사","inventors":["김상윤","한창우","이도균","신민규","유종욱","이재원"],"publication_date":"2026-05-21","filing_date":"2026-05-06","priority_date":"2018-10-18","cpc_codes":["G","G06","G06F","G06F3/00","G06F3/16","G06F3/167","G","G06","G06F","G06F3/00","G06F3/01","G06F3/03","G06F3/041","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0487","G06F3/0488","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G10","G10L","G10L15/00","G10L15/22","G","G10","G10L","G10L15/00","G10L15/22","G10L2015/221"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260073329A/en"},{"publication_number":"KR20260073322A","title":"Photonic qubit-based quantum computing device integrated on a semiconductor substrate","abstract":"본 발명은 반도체 기판(100) 상에 집적된 광자 큐비트 기반 양자 컴퓨팅 장치를 개시한다. 질화규소(Si₃N₄) 마이크로링 공명기(200)에서 대칭 이탈 조정된 이중 펌프 자발적 사광자 혼합(SFWM)과 양자 펌프 고갈 체제를 통해 비가우시안 슈뢰딩거 고양이 유사 상태의 광자 큐비트가 생성된다. 마이크로링 공명기(200)에서 수직 방향으로 연장되는 광도파로(300)는 S자형 굴곡 구간(320)을 통해 수평 방향의 트렌치 구조(400) 내에서 평행하게 배치되어 소실파 결합에 의한 양자 얽힘을 수행한다. 트렌치 구조(400)의 측벽(410) 및 바닥면(420)에는 ALD 공정으로 형성된 TiN 또는 W 광차단 클래딩층(430)이 구비되어 광 누설 및 크로스토크를 차단한다. 초전도 나노와이어 단일 광자 검출기(SNSPD, 710), 호모다인 검출 회로(740), 및 양자 상태 토모그래피 처리부(760)가 동일 기판 위에 집적되어 완전 온칩 양자 측정을 구현한다. 직렬 트렌치 구조 기반 마흐-젠더 간섭계 구조로 2큐비트 CZ/CNOT 게이트를 구현하며, 병렬 세트와 광 스위치 어레이(900) 및 광 지연선(910)으로 확률적 연산 성공률을 향상시킨다. The present invention discloses a photon qubit-based quantum computing device integrated on a semiconductor substrate (100). Photon qubits in a non-Gaussian Schrödinger cat-like state are generated through a dual pump spontaneous four-photon mixing (SFWM) and a quantum pump depletion regime in a silicon nitride (Si₃N₄) micro-ring resonator (200) that is symmetrically tuned. An optical waveguide (300) extending vertically from the micro-ring resonator (200) is arranged parallel within a horizontal trench structure (400) through an S-shaped bend section (320) to perform quantum entanglement by disappearing wave coupling. A TiN or W light-blocking cladding layer (430) formed by an ALD process is provided on the sidewalls (410) and bottom surface (420) of the trench structure (400) to block light leakage and crosstalk. A superconducting nanowire single-photon detector (SNSPD, 710), a homodyne detection circuit (740), and a quantum state tomography processing unit (760) are integrated on the same substrate to enable full on-chip quantum measurement. A 2-qubit CZ/CNOT gate is implemented using a serial trench structure-based Mach-Zehnder interferometer structure, and the success rate of probabilistic computation is improved using a parallel set, an optical switch array (900), and an optical delay line (910).","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-21","filing_date":"2026-05-01","priority_date":"","cpc_codes":["H","H10","H10N","H10N69/00","G","G02","G02B","G02B6/00","G02B6/10","G02B6/12","G02B6/12007","G","G02","G02B","G02B6/00","G02B6/10","G02B6/12","G02B6/122","G","G06","G06N","G06N10/00","G06N10/40","G","G06","G06N","G06N10/00","G06N10/70"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260073322A/en"},{"publication_number":"KR20260072558A","title":"Noise-Prediction-Based Adaptive Quantum Circuit Reconfiguration System and Method","abstract":"본 발명은 양자 컴퓨팅 환경에서 발생하는 시간적 및 공간적 노이즈 변동을 예측하고, 예측 결과에 따라 양자 회로의 실행 구조를 가변적으로 재구성하는 시스템 및 방법에 관한 것이다. 본 발명은 오류증후군 이력, 교정 드리프트, T1/T2 결맞음 데이터, 판독 혼동 행렬, 혼선 행렬, 환경 로그 및 양자 처리 장치 연결성 그래프 정보를 수집하는 데이터 수집부와, 상기 데이터를 이용하여 오류 가능성이 높은 물리 자원 또는 회로 구간을 확률 분포 및 신뢰구간으로 예측하는 노이즈 예측 엔진을 포함한다. 본 발명의 핵심 구성은 (1) 교환 게이트 추가 비용, 예측 오류율 이익 및 예상 수명 점수를 함께 고려하여 최적 대체 큐비트를 선택하는 위상 인식 예측 재매핑(TAPR), (2) 돌발 잡음 또는 이상 이벤트를 감지하고 영향 큐비트를 격리한 후 논리적 우회 연결을 포함하여 회로를 재편성하는 시간적 노이즈 버스트 격리(TNBI), (3) 예측 신뢰구간과 추가 자원 비용을 함께 고려하여 자원 투입 수준을 단계적으로 결정하는 신뢰도 게이트 자원 에스컬레이션(CGRE), 및 (4) 특정 시간대의 노이즈 집중도를 반영하여 게이트 실행 시점을 미세 조정하는 리소스 가중치 기반 동적 게이트 스케줄링(RDGS)을 포함한다. 본 발명은 계층적 노이즈 적응형 제어 시스템과 독립적으로 또는 직교적으로 결합되는 상위 회로 재구성 계층으로서, 실행 제어 파라미터의 단순 최적화가 아니라 노이즈 예측 결과에 따라 회로의 매핑·스케줄·깊이 및 실행 자원 배치를 변경하는 상위 회로 재구성 계층을 직접 대상으로 한다. The present invention relates to a system and method for predicting temporal and spatial noise fluctuations occurring in a quantum computing environment and variably reconfiguring the execution structure of a quantum circuit according to the prediction results. The present invention includes a data collection unit that collects error syndrome history, correction drift, T1/T2 coherence data, read confusion matrix, crosstalk matrix, environment log, and quantum processing device connectivity graph information, and a noise prediction engine that uses the data to predict physical resources or circuit sections with a high probability of error as probability distributions and confidence intervals. The core components of the present invention include (1) phase-aware predictive remapping (TAPR) that selects an optimal replacement qubit by considering the additional cost of the exchange gate, the profit from the predicted error rate, and the expected lifespan score together; (2) temporal noise burst isolation (TNBI) that detects sudden noise or abnormal events, isolates the affected qubit, and then reorganizes the circuit including logical bypass connections; (3) confidence gate resource escalation (CGRE) that determines the level of resource input in stages by considering the prediction confidence interval and the additional resource cost together; and (4) resource-weighted dynamic gate scheduling (RDGS) that fine-tunes the timing of gate execution by reflecting the noise concentration of a specific time period. The present invention directly targets a higher-level circuit reconfiguration layer that is coupled independently or orthogonally to a hierarchical noise-adaptive control system, which modifies the mapping, scheduling, depth, and execution resource allocation of the circuit according to noise prediction results, rather than simply optimizing execution control parameters.","assignee":"아이이엔건축사사무소주식회사","inventors":["정찬희"],"publication_date":"2026-05-20","filing_date":"2026-05-03","priority_date":"","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/70","G","G06","G06N","G06N10/00","G06N10/20","G","G06","G06N","G06N10/00","G06N10/40","B","B82","B82Y","B82Y10/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260072558A/en"},{"publication_number":"KR20260071173A","title":"Artificial intelligence cooking appliance and operating method for the same","abstract":"본 개시의 실시 예들은 인공 지능을 이용한 조리 기기 및 그 동작 방법에 관한 것으로, 인공 지능 조리 기기는 조리 대상 제품을 조리하기 위한 구동 장치, 상기 조리 대상 제품과 연관된 이미지를 획득하는 이미지 센서 및 상기 구동 장치 및 상기 이미지 센서와 작동적으로 연결된 적어도 하나의 프로세서를 포함하고, 상기 적어도 하나의 프로세서는, 상기 조리 기기가, 상기 조리 대상 제품과 연관된 이미지를 획득하고, 상기 이미지에 기초하여 상기 조리 대상 제품의 조리 정보를 획득하고, 상기 조리 정보에 기초하여 상기 조리 대상 제품의 조리를 수행하도록 제어할 수 있다. Embodiments of the present disclosure relate to a cooking device using artificial intelligence and a method of operating the same. The artificial intelligence cooking device includes a driving device for cooking a product to be cooked, an image sensor for acquiring an image associated with the product to be cooked, and at least one processor operatively connected to the driving device and the image sensor. The at least one processor can control the cooking device to acquire an image associated with the product to be cooked, acquire cooking information of the product to be cooked based on the image, and perform cooking of the product to be cooked based on the cooking information.","assignee":"엘지전자 주식회사","inventors":["김보은"],"publication_date":"2026-05-19","filing_date":"2026-05-12","priority_date":"","cpc_codes":["H","H05","H05B","H05B6/00","H05B6/64","H05B6/6435","G","G06","G06N","G06N20/00","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/60","G06V20/68","G","G06","G06V","G06V30/00","G06V30/10","G06V30/14","G06V30/148","G06V30/153","G","G06","G06V","G06V30/00","G06V30/10","G06V30/22","G06V30/224","H","H05","H05B","H05B6/00","H05B6/64","H05B6/6447","H","H05","H05B","H05B6/00","H05B6/64","H05B6/66","H05B6/668","G","G06","G06V","G06V2201/00","G06V2201/09"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260071173A/en"},{"publication_number":"KR20260071170A","title":"Technologies for providing a scalable architecture for performing compute operations in memory","abstract":"메모리에서 계산 동작들을 효율적으로 수행하기 위한 스케일링가능 아키텍처를 제공하는 기술들은 메모리 매체에 결합된 매체 액세스 회로를 갖는 메모리를 포함한다. 매체 액세스 회로는, 요청된 동작을 수행하기 위해 메모리 매체로부터의 데이터에 액세스하고, 매체 액세스 회로에 포함된 다수의 계산 로직 유닛들 각각을 사용하여, 액세스된 데이터에 대해 요청된 동작을 동시에 수행하고, 요청된 동작의 실행으로부터 생성된 결과 데이터를 메모리 매체에 기입한다. Technologies providing a scalable architecture for efficiently performing computational operations in memory include a memory having a media access circuit coupled to a memory medium. The media access circuit accesses data from the memory medium to perform a requested operation, simultaneously performs the requested operation on the accessed data using each of a plurality of computational logic units included in the media access circuit, and writes the result data generated from the execution of the requested operation to the memory medium.","assignee":"인텔 코포레이션","inventors":["시게끼 토미시마","스리칸쓰 스리니바산","체탄 차우한","라제쉬 선다람","재와드 비. 칸"],"publication_date":"2026-05-19","filing_date":"2026-05-06","priority_date":"2019-03-29","cpc_codes":["G","G06","G06F","G06F15/00","G06F15/76","G06F15/78","G06F15/7867","G","G06","G06F","G06F12/00","G06F12/02","G06F12/0207","G","G06","G06F","G06F15/00","G06F15/76","G06F15/78","G06F15/7807","G06F15/7821","G","G06","G06F","G06F15/00","G06F15/76","G06F15/78","G06F15/7839","G06F15/7842","G","G06","G06F","G06F15/00","G06F15/76","G06F15/80","G06F15/8007","G06F15/803","G","G06","G06F","G06F17/00","G06F17/10","G06F17/16","G","G06","G06F","G06F3/00","G06F3/06","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G11","G11C","G11C5/00","G11C5/02","G11C5/04"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260071170A/en"},{"publication_number":"KR20260070373A","title":"Apparatus and method for AI model weight reduction using pseudo-weight based contribution testing","abstract":"본 발명은 의사 가중치(Pseudo Weight) 기반 기여도 테스트를 이용한 인공지능 모델 경량화 장치 및 방법에 관한 것이다. 본 발명에 따른 인공지능 모델 경량화 장치(100)는, 인공지능 모델(200)의 개별 가중치 또는 뉴런 그룹을 식별하는 대상 식별부(110); 식별된 대상의 실제 가중치(241)와 통계적 특성은 유사하되 연결 정보가 파괴된 의사 가중치(242)를 생성하는 의사 데이터 생성부(120); 의사 가중치(242) 치환 전후의 추론 성능 변동으로부터 기여도 점수를 산출하는 민감도 분석부(130); 및 기여도 점수가 임계치 미만인 가중치 그룹을 영구 삭제하거나 영(Zero)으로 고정하는 가지치기 수행부(140)를 포함한다. 본 발명은 학습 데이터 없이 동적 반응 테스트만으로 가중치의 실질적 기여도를 정밀하게 측정하여, 90% 이상의 파라미터를 제거하면서도 5% 이내의 성능 열화를 유지하는 탁월한 경량화 효과를 제공한다. The present invention relates to an apparatus and method for lightweighting an artificial intelligence model using a pseudo-weight-based contribution test. An artificial intelligence model lightweighting apparatus (100) according to the present invention comprises: a target identification unit (110) for identifying individual weights or neuron groups of an artificial intelligence model (200); a pseudo-data generation unit (120) for generating pseudo-weights (242) that have statistical characteristics similar to the actual weights (241) of the identified targets but have destroyed connection information; a sensitivity analysis unit (130) for calculating a contribution score from the fluctuation in inference performance before and after the substitution of pseudo-weights (242); and a pruning execution unit (140) for permanently deleting weight groups whose contribution scores are below a threshold or fixing them to zero. The present invention provides an excellent lightweighting effect by precisely measuring the actual contribution of weights using only dynamic response tests without training data, thereby removing more than 90% of parameters while maintaining performance degradation within 5%.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-19","filing_date":"2026-04-30","priority_date":"","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260070373A/en"},{"publication_number":"KR20260070374A","title":"Apparatus and method for neural network training to prevent overfitting and enhance robustness using pseudo-weights","abstract":"본 발명은 신경망 학습 장치 및 신경망 강건성 향상 방법에 관한 것이다. 본 발명의 장치는, 학습 대상 신경망 모델의 실제 가중치를 저장하는 가중치 저장부, 실제 가중치의 통계적 특성(평균·분산)을 기반으로 정규 분포를 따르는 의사 가중치를 생성하는 의사 데이터 생성부, 학습 반복마다 무작위로 선택된 일부 가중치를 의사 가중치로 일시 치환하는 강건성 제어부, 혼합 상태에서 손실 함수를 최소화하며 실제 가중치를 업데이트하는 모델 학습부, 및 학습 완료 후 의사 가중치를 제거하고 추론에 최적화된 실제 가중치를 복원하는 추론 전환부를 포함한다. 레이어별 차등 치환 확률과 통계적 일관성 유지 메커니즘으로 과적합을 방지하고 손실 공간의 평탄한 최솟값 탐색을 유도한다. The present invention relates to a neural network learning device and a method for improving neural network robustness. The device of the present invention includes a weight storage unit that stores actual weights of a neural network model to be trained; a pseudo-data generation unit that generates pseudo-weights following a normal distribution based on the statistical characteristics (mean and variance) of the actual weights; a robustness control unit that temporarily replaces some randomly selected weights with pseudo-weights for each training iteration; a model training unit that updates actual weights while minimizing the loss function in a mixed state; and an inference transition unit that removes the pseudo-weights and restores the actual weights optimized for inference after training is completed. Overfitting is prevented and the search for a flat minimum value in the loss space is induced through layer-specific differential substitution probabilities and a statistical consistency maintenance mechanism.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-19","filing_date":"2026-04-30","priority_date":"","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06F","G06F17/00","G06F17/10","G06F17/18","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260070374A/en"},{"publication_number":"KR20260070380A","title":"Rotary Residency Loading System and Method for Large Language Models on Memory-Constrained Accelerators","abstract":"본 발명은 전체 가중치 footprint가 가속기 메모리 용량을 초과하는 대규모 언어 모델을 단일 가속기 또는 소수의 가속기 상에서 실행하기 위한 로터리 회전 적재 시스템에 관한 것이다. 본 발명의 시스템은 호스트 메모리 또는 저장장치 계층에 신경망 서브모듈 가중치를 분산 저장하는 호스트 측 가중치 저장부, 입력 토큰 시퀀스의 진행에 따라 슬롯 내 서브모듈이 cyclical 순방향 회전 또는 역방향 회전에 의해 갱신되는 가속기 메모리 내 로터리 회전 슬롯 그룹, 서브모듈 식별자를 슬롯 위치에 매핑하는 가속기 측 lookup table, 입력 hidden state, routing vector 또는 routing trajectory에 대한 회전 변환의 결과로서 산출되는 회전 사영에 기초하여 로터리 회전을 제어하는 회전 제어부를 포함한다. 상기 cyclical 회전은 단순 사용 빈도 기반 강등 또는 임계값 기반 분류와 별도로 회전 변환의 결과에 의해 결정된다. 본 발명에 따르면, 모델 footprint가 가속기 메모리를 초과하는 경우에도 단일 가속기 상에서 토큰 생성을 수행할 수 있고, 종래 LRU 대비 의미 흐름 추종성이 향상되며, host-device 동기화 비용이 감소되고, 폐쇄망 및 온프레미스 운용이 가능하다 The present invention relates to a rotary loading system for running a large-scale language model, in which the total weight footprint exceeds the accelerator memory capacity, on a single accelerator or a small number of accelerators. The system of the present invention comprises a host-side weight storage unit that distributes and stores neural network submodule weights in a host memory or storage layer, a rotary rotation slot group in accelerator memory in which submodules within the slot are updated by cyclical forward rotation or reverse rotation according to the progress of an input token sequence, an accelerator-side lookup table that maps submodule identifiers to slot positions, and a rotation control unit that controls rotary rotation based on a rotation projection calculated as a result of a rotation transformation on an input hidden state, routing vector, or routing trajectory. The cyclical rotation is determined by the result of the rotation transformation separately from simple usage frequency-based demotion or threshold-based classification. According to the present invention, token generation can be performed on a single accelerator even when the model footprint exceeds the accelerator memory, semantic flow following is improved compared to conventional LRU, host-device synchronization costs are reduced, and closed network and on-premises operation is possible.","assignee":"조명준","inventors":["조명준"],"publication_date":"2026-05-19","filing_date":"2026-04-30","priority_date":"","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06F","G06F17/00","G06F17/10","G06F17/18","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260070380A/en"},{"publication_number":"KR20260070099A","title":"Apparatus for controlling refrigerator","abstract":"본 개시는 환경 데이터에 기초하여 냉장고 내 식품의 부패 확률을 예측하고, 예측 결과에 따라 급속 냉장을 수행할 수 있는 냉장고 제어 장치에 관한 것이다. 본 개시의 일 관점에 의한 냉장고 제어 장치는 냉장고에 탑재된 제어 장치에 있어서, 냉장고 내부의 온도 및 습도를 포함하는 환경 데이터를 시간의 흐름에 따라 획득하는 입력부; 프로세서; 및 상기 프로세서에 의해 실행되며 상기 환경 데이터를 포함하는 시퀀스 데이터를 입력받아 식품의 부패 확률을 출력하도록 학습된 인공지능 모델을 포함하고, 상기 프로세서는 상기 인공지능 모델로부터 출력된 부패 확률이 설정된 임계값보다 큰 경우 냉장고의 냉각 동작을 강화하도록 냉장고를 제어할 수 있다. The present disclosure relates to a refrigerator control device capable of predicting the probability of spoilage of food inside a refrigerator based on environmental data and performing rapid cooling according to the prediction result. A refrigerator control device according to one aspect of the present disclosure comprises a control device mounted on a refrigerator, wherein the control device comprises: an input unit that acquires environmental data including the temperature and humidity inside the refrigerator over time; a processor; and an artificial intelligence model that is executed by the processor and is trained to receive sequence data including the environmental data and output a probability of spoilage of food, wherein the processor can control the refrigerator to intensify the cooling operation of the refrigerator when the probability of spoilage output from the artificial intelligence model is greater than a set threshold.","assignee":"엘지전자 주식회사","inventors":["김성애"],"publication_date":"2026-05-18","filing_date":"2026-05-11","priority_date":"","cpc_codes":["F","F25","F25D","F25D29/00","F25D29/005","F","F25","F25D","F25D29/00","F25D29/008","G","G05","G05B","G05B13/00","G05B13/02","G05B13/0265","G05B13/027","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G05","G05B","G05B2219/00","G05B2219/20","G05B2219/26","G05B2219/2654"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260070099A/en"},{"publication_number":"KR20260070089A","title":"System for Predicting Renewable Energy Power Generation using Data from Multiple Weather Stations","abstract":"본 발명의 다중 기상관측소 데이터를 이용한 신재생에너지 발전량 예측 방법은, 대상 평면 지역을 기상관측소를 꼭지점으로 하는 다수개의 삼각형 영역들로 분할하는 단계; 분할된 각 삼각형 영역에 신재생에너지 발전소들을 클러스터링하는 단계; 각 신재생에너지 발전소의 발전량을 예측하기 위해, 상기 발전소가 클러스터링된 삼각형 영역의 꼭지점들에 위치한 기상관측소들을 식별하는 단계; 및 예측 모델에 식별된 기상관측소들의 기상 정보들을 적용하여 상기 발전소의 발전량을 예측하는 단계를 포함할 수 있다. The method for predicting renewable energy generation using multiple weather station data according to the present invention may include: a step of dividing a target planar area into a plurality of triangular regions with weather stations as vertices; a step of clustering renewable energy power plants in each divided triangular region; a step of identifying weather stations located at the vertices of the triangular regions where the power plants are clustered in order to predict the generation amount of each renewable energy power plant; and a step of predicting the generation amount of the power plant by applying weather information of the identified weather stations to a prediction model.","assignee":"한국전력공사","inventors":["문종희","권성철","김홍석","정재익","송근주"],"publication_date":"2026-05-18","filing_date":"2026-04-30","priority_date":"","cpc_codes":["H","H02","H02S","H02S50/00","G","G01","G01W","G01W1/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","Y","Y02","Y02E","Y02E10/00","Y02E10/50"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260070089A/en"},{"publication_number":"KR20260070086A","title":"Method, Apparatus, and Computer-Readable Recording Medium for 5-ary Weight Conversion of Floating-Point Based Large Language Models","abstract":"본 발명은 기학습된 부동소수점(Floating Point) 기반 거대언어모델(LLM)의 가중치를 변환하는 방법, 장치 및 컴퓨터 판독 가능 기록 매체에 관한 것이다. 본 발명은 거대언어모델의 가중치 행렬을 구성하는 각 가중치 원소를 5개의 레벨 {-1, -0.5, 0, 0.5, 1} 중 어느 하나로 매핑하는 5진 양자화 단계 및 5진 양자화된 가중치를 이용하여 거대언어모델을 재구성하는 모델 재구성 단계를 포함한다. 바람직한 실시예에 따르면, 제1 임계값(Δ₁) 및 제2 임계값(Δ₂)을 계층별 가중치 분포의 절대값 평균으로부터 산출하여, |W| < Δ₁이면 0, Δ₁ ≤ |W| < Δ₂이면 sgn(W)×0.5, |W| ≥ Δ₂이면 sgn(W)×1로 각 원소를 매핑한다. 본 발명에 따르면, 부동소수점 대비 메모리가 약 85% 절감되고, 삼진({-1, 0, +1}) 양자화 대비 표현력이 향상되어 퍼플렉시티 성능이 개선되며, ±0.5 레벨의 1비트 시프트 연산 특성으로 인해 행렬 연산에서 부동소수점 곱셈이 실질적으로 제거되는 효과를 달성한다. The present invention relates to a method, apparatus, and computer-readable recording medium for transforming weights of a pre-trained floating-point-based large language model (LLM). The present invention includes a pentatonic quantization step for mapping each weight element constituting the weight matrix of the large language model to any one of five levels {-1, -0.5, 0, 0.5, 1}, and a model reconstruction step for reconstructing the large language model using the pentatonic quantized weights. According to a preferred embodiment, a first threshold value (Δ₁) and a second threshold value (Δ₂) are calculated from the absolute mean of the weight distribution by hierarchy, and each element is mapped to 0 if |W| < Δ₁, sgn(W)×0.5 if Δ₁ ≤ |W| < Δ₂, and sgn(W)×1 if |W| ≥ Δ₂. According to the present invention, memory is reduced by approximately 85% compared to floating-point, perplexity performance is improved by enhancing expressiveness compared to ternary ({-1, 0, +1}) quantization, and floating-point multiplication in matrix operations is substantially eliminated due to the ±0.5 level 1-bit shift operation characteristic.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-18","filing_date":"2026-04-29","priority_date":"","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/483","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260070086A/en"},{"publication_number":"KR20260070084A","title":"Reliability-enhanced artificial intelligence computing device based on real-time hardware fault detection using diagnostic expert weights and method for the same","abstract":"본 발명은 인공지능 연산 장치 및 그 하드웨어 결함 자가진단 방법에 관한 것이다. 본 발명의 인공지능 연산 장치는, 미리 연산 결과값이 확정된 진단용 의사 가중치를 관리하는 진단 데이터 관리부, 실제 추론 연산과 의사 가중치를 이용한 진단 연산을 병렬 또는 교차로 수행하는 연산 수행부, 진단 연산의 출력값과 기준값을 실시간 비교하여 하드웨어 논리 오류를 판별하는 결함 분석부, 및 오류 감지 시 신뢰성 등급을 조정하거나 재연산을 명령하는 신뢰성 제어부를 포함한다. 진단 연산은 저정밀도 의사 가중치를 사용하여 연산 오버헤드를 5% 이내로 유지하며, 온도·전압 센서와 연동한 환경 적응형 샘플링으로 진단 정밀도를 동적으로 강화하고, 다중 코어 환경에서 결함 코어를 격리하여Fail-Operational 특성을 실현한다. The present invention relates to an artificial intelligence computing device and a method for self-diagnosing hardware defects thereof. The artificial intelligence computing device of the present invention includes a diagnostic data management unit that manages diagnostic pseudo-weights in which the computation result value is determined in advance; a computation execution unit that performs actual inference computations and diagnostic computations using pseudo-weights in parallel or alternately; a defect analysis unit that determines hardware logic errors by comparing the output value of the diagnostic computation with a reference value in real time; and a reliability control unit that adjusts the reliability grade or commands re-computation upon error detection. The diagnostic computation maintains computational overhead within 5% by using low-precision pseudo-weights, dynamically enhances diagnostic precision through environment-adaptive sampling linked with temperature and voltage sensors, and realizes fail-operational characteristics by isolating defective cores in a multi-core environment.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-18","filing_date":"2026-04-29","priority_date":"","cpc_codes":["G","G06","G06F","G06F11/00","G06F11/22","G06F11/2263","G","G06","G06F","G06F21/00","G06F21/60","G06F21/64","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260070084A/en"},{"publication_number":"KR20260069503A","title":"Method for generating document to test of user","abstract":"본 개시의 몇몇 실시예에 따른, 컴퓨팅 장치에 의해 수행되는, 사용자를 테스트하기 위한 문서를 생성하는 방법으로서, 상기 방법은: 외부 기기로부터 사용자의 식별정보 및 테스트의 종류에 관한 정보를 수신하는 단계; 사전 학습된 제 1 신경망 모델을 이용하여, 상기 테스트의 종류에 관한 정보에 대응되는 복수의 후보 문항들을 획득하는 단계; 상기 사용자의 식별정보가 라벨링된 사전 저장된 복수의 원문항들에 기초하여, 상기 사용자의 식별정보에 대응되는 적어도 하나의 사전 응답 문항을 획득하는 단계; 상기 복수의 후보 문항들 중에서 상기 적어도 하나의 사전 응답 문항에 대응되는 문항을 제거하여 최종 문항을 획득하는 단계; 및 상기 최종 문항을 포함하는 제 1 문서를 생성하여 상기 외부 기기로 전송하는 단계;를 포함할 수 있다. A method for generating a document for testing a user, performed by a computing device according to some embodiments of the present disclosure, the method may include: receiving information regarding user identification information and the type of test from an external device; obtaining a plurality of candidate questions corresponding to the information regarding the type of test using a pre-trained first neural network model; obtaining at least one pre-response question corresponding to the user identification information based on a plurality of pre-stored original questions labeled with the user identification information; obtaining a final question by removing a question corresponding to the at least one pre-response question from among the plurality of candidate questions; and generating a first document including the final question and transmitting it to the external device.","assignee":"삼성생명보험주식회사","inventors":["김정동","조용진","임진성"],"publication_date":"2026-05-15","filing_date":"2026-04-27","priority_date":"","cpc_codes":["G","G16","G16H","G16H10/00","G16H10/20","A","A61","A61B","A61B5/00","A61B5/16","A61B5/165","G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G16","G16H","G16H15/00","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/70","G","G16","G16H","G16H70/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260069503A/en"},{"publication_number":"KR20260069502A","title":"Artificial intelligence system and method for editing image based on relation between objects","abstract":"프로세서, 및 인스트럭션들을 저장하는 메모리를 포함하며, 상기 프로세서는 상기 인스트럭션들을 실행함으로써, 원본 이미지를 수신하고, 상기 원본 이미지에서 복수의 객체들을 인식하여, 상기 복수의 객체들을 나타내는 객체 정보를 생성하고, 상기 원본 이미지 및 상기 객체 정보에 기초하여 상기 복수의 객체들 간의 관계를 나타내는 객체 관계 그래프를 생성하고, 상기 객체 관계 그래프를 이미지 수정 그래프 뉴럴 네트워크(Graph Neural Network, GNN) 모델에 입력함으로써 상기 복수의 객체들에 각각 적용될 이미지 효과들을 포함하는 이미지 효과 데이터를 획득하고, 상기 원본 이미지, 상기 객체 정보, 및 상기 이미지 효과 데이터에 기초하여 수정 이미지를 생성하는, 이미지 수정 시스템이 개시된다. An image modification system is disclosed, comprising a processor and a memory for storing instructions, wherein the processor receives an original image by executing the instructions, recognizes a plurality of objects in the original image, generates object information representing the plurality of objects, generates an object relationship graph representing the relationship between the plurality of objects based on the original image and the object information, obtains image effect data including image effects to be applied to each of the plurality of objects by inputting the object relationship graph into an image modification graph neural network (GNN) model, and generates a modified image based on the original image, the object information, and the image effect data.","assignee":"삼성전자주식회사","inventors":["서찬원","김은서","이영은","이홍표"],"publication_date":"2026-05-15","filing_date":"2026-04-24","priority_date":"","cpc_codes":["G","G06","G06T","G06T11/00","G06T11/60","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T11/00","G06T11/20","G06T11/26","G","G06","G06T","G06T7/00","G06T7/10","G06T7/13","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260069502A/en"},{"publication_number":"KR20260068741A","title":"Device and method for detecting abnormality of motor, and non-transitory computer-readable storage medium storing program for performing the method","abstract":"모터 이상 감지 장치 및 방법, 상기 방법의 수행을 위한 프로그램이 저장된 비일시적 컴퓨터 판독 가능 저장 매체가 개시된다. 본 발명의 일 측면에 따른 모터 이상 감지 장치는, 차량의 EPS(Electric Power Steering) 시스템에서 조향 보조 토크를 생성하는 모터의 이상을 감지하는 모터 이상 감지 장치로서, 메모리; 및 프로세서;를 포함하고, 상기 메모리는 상기 차량의 상태와 관련된 상태 데이터, 상기 차량의 운전자의 스티어링 조작과 관련된 조작 데이터를 입력 받아 상기 모터의 출력과 관련된 물리량을 추정하는 인공 신경망 모델을 저장하고, 상기 프로세서는 상기 상태 데이터 및 상기 조작 데이터를 상기 인공 신경망 모델에 입력하고 연산을 수행하여 상기 물리량의 추정치를 출력하고, 상기 추정치를 상기 물리량의 실측치와 비교하여 상기 모터의 이상 여부를 감지할 수 있다. A motor abnormality detection device and method, and a non-transient computer-readable storage medium storing a program for performing the method are disclosed. A motor abnormality detection device according to one aspect of the present invention is a motor abnormality detection device for detecting an abnormality in a motor that generates steering assist torque in an Electric Power Steering (EPS) system of a vehicle, and comprises: a memory; and a processor; wherein the memory stores an artificial neural network model that receives state data related to the state of the vehicle and operation data related to the steering operation of the driver of the vehicle, and estimates a physical quantity related to the output of the motor; and the processor inputs the state data and the operation data into the artificial neural network model and performs calculations to output an estimate of the physical quantity, and can detect whether there is an abnormality in the motor by comparing the estimate with an actual value of the physical quantity.","assignee":"에이치엘만도 주식회사","inventors":["이도헌","김규원","원종익"],"publication_date":"2026-05-14","filing_date":"2026-05-04","priority_date":"","cpc_codes":["B","B62","B62D","B62D5/00","B62D5/04","B62D5/0457","B62D5/0481","B62D5/0487","B","B60","B60R","B60R16/00","B60R16/02","B60R16/023","B60R16/0231","B60R16/0232","G","G01","G01R","G01R31/00","G01R31/005","G01R31/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G07","G07C","G07C5/00","H","H04","H04L","H04L12/00","H04L12/28","H04L12/46","B","B60","B60Y","B60Y2306/00","B60Y2306/15"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260068741A/en"},{"publication_number":"AU2026203196A1","title":"Systems and methods for detecting occupancy using radio signals","abstract":"A sensor system for determining an occupancy state in a space, generally including: one or more radio-capable devices disposed in the space, wherein at least one of the one or more radio-capable devices is configured to propagate radio signals through the space and at least one of the one or more radio-capable devices is configured to receive radio signals that have traversed at least a portion of the space; and at least one processor configured to execute an occupancy-centric algorithm to: (a) determine an occupancy state of the space based on disturbances, interference, attenuation, or other changes in the propagated radio signals that are associated with a presence, movement, or absence of at least one person in the space; and (b) output an occupancy-derived control signal to cause performance of an automation action by at least one building, home, or room system associated with the space in response to the determined occupancy state.","assignee":"Strong Force VCN Portfolio 2019 LLC","inventors":["Charles Cella","Stephen ELIAS","Eric Giler","Katherine HALL"],"publication_date":"2026-05-14","filing_date":"2026-04-29","priority_date":"2019-07-08","cpc_codes":["G","G01","G01S","G01S7/00","G01S7/003","G01S7/006","H","H04","H04W","H04W4/00","H04W4/30","H04W4/38","F","F24","F24F","F24F11/00","F24F11/30","F24F11/49","G","G01","G01S","G01S13/00","G01S13/003","G","G01","G01S","G01S13/00","G01S13/02","G01S13/50","G01S13/52","G01S13/56","G","G05","G05B","G05B15/00","G05B15/02","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06N","G06N20/00","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06312","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06315","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06398","G","G06","G06Q","G06Q10/00","G06Q10/08","G","G16","G16H","G16H40/00","G16H40/60","G16H40/67","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/70","H","H04","H04B","H04B17/00","H04B17/30","H","H04","H04W","H04W24/00","F","F24","F24F","F24F2120/00","F24F2120/10","F24F2120/12","F","F24","F24F","F24F2120/00","F24F2120/10","F24F2120/14"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026203196A1/en"},{"publication_number":"AU2026203234A1","title":"Cross-media measurement device and method","abstract":"CROSS-MEDIA MEASUREMENT DEVICE AND METHOD A method of identifying media content presented on a display device includes determining a selected input source providing a video signal to the display device, and then selecting a first set of content identification rules when it is determined that the selected input source is a first input source, and selecting a second set of content identification rules when it is determined that the selected input source is a second input source. The method further comprises applying the selected first set or second set of content identification rules to the video signal in order to generate content identification data for the media content presented on the display device. Application of the content identification rules includes waiting for a trigger event and applying an algorithm to one or more frames of the video signal following the trigger event. CROSS-MEDIA MEASUREMENT DEVICE AND METHOD","assignee":"Hyphametrics Inc","inventors":["Joanna DREWS","Gerardo Lopez Zamudio"],"publication_date":"2026-05-14","filing_date":"2026-04-29","priority_date":"2019-07-09","cpc_codes":["H","H04","H04N","H04N21/00","H04N21/40","H04N21/43","H04N21/44","H04N21/44008","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V20/00","G06V20/40","G06V20/46","G","G06","G06V","G06V20/00","G06V20/40","G06V20/48","G","G06","G06V","G06V20/00","G06V20/40","G06V20/49","H","H04","H04N","H04N21/00","H04N21/20","H04N21/25","H04N21/254","H","H04","H04N","H04N21/00","H04N21/20","H04N21/25","H04N21/258","H04N21/25866","H","H04","H04N","H04N21/00","H04N21/20","H04N21/25","H04N21/258","H04N21/25866","H04N21/25883","H","H04","H04N","H04N21/00","H04N21/40","H04N21/43","H04N21/44","H04N21/4402","H","H04","H04N","H04N21/00","H04N21/40","H04N21/43","H04N21/442","H04N21/44213","H04N21/44218","H","H04","H04N","H04N21/00","H04N21/40","H04N21/43","H04N21/442","H04N21/44213","H04N21/44222","H","H04","H04N","H04N21/00","H04N21/40","H04N21/43","H04N21/442","H04N21/44227","H","H04","H04N","H04N21/00","H04N21/40","H04N21/45","H04N21/4508","H04N21/4516"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026203234A1/en"},{"publication_number":"AU2026203135A1","title":"Selection-inference neural network systems","abstract":"Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating a response to a query input using a selection- inference neural network.","assignee":"GDM Holding LLC","inventors":["Antonia Phoebe Nina CRESWELL"],"publication_date":"2026-05-14","filing_date":"2026-04-27","priority_date":"2022-05-13","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06F","G06F40/00","G06F40/30","G06F40/35","G","G06","G06N","G06N3/00","G06N3/004","G06N3/008","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N5/00","G06N5/04","G06N5/045","G","G06","G06N","G06N5/00","G06N5/04","G06N5/046"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026203135A1/en"},{"publication_number":"KR20260068731A","title":"Semiconductor Quantum Dot-Based Quantum Computing Device Having a Staggered Gate Architecture","abstract":"본 발명은 반도체 양자점 기반 양자 컴퓨팅 장치에 관한 것으로, 단일 전자를 구속하는 양자점이 형성되는 채널 영역을 포함하는 기판, 터널링 장벽을 형성하는 복수의 배리어 게이트를 포함하는 제1 전극층, 절연층, 및 양자점의 전위를 조절하는 복수의 플런저 게이트를 포함하는 제2 전극층을 구비한다. 제1 전극층 내의 인접한 배리어 게이트들은 수평 방향으로 서로 엇갈리게(Staggered) 배열되고, 제2 전극층 내의 인접한 플런저 게이트들 또한 수평 방향으로 서로 엇갈리게 배열된다. 이 스테거드 배열에 의해 동일 전극층 내 인접 동종 게이트들 간의 실효 이격 거리가 증대되고 상호 커패시턴스가 감소하여, 층내 수평 방향 정전기적 크로스토크가 구조적으로 최소화된다. 이로써 배리어 게이트의 터널 결합 독립 제어 정밀도와 플런저 게이트의 양자점 포텐셜 독립 제어 정밀도가 향상되어, 단일 큐비트 및 2큐비트 게이트 충실도가 개선되고 코히런스 시간이 연장된다. The present invention relates to a semiconductor quantum dot-based quantum computing device comprising a substrate including a channel region in which a quantum dot confining a single electron is formed, a first electrode layer including a plurality of barrier gates forming a tunneling barrier, an insulating layer, and a second electrode layer including a plurality of plunger gates controlling the potential of the quantum dot. Adjacent barrier gates within the first electrode layer are arranged staggered relative to each other in the horizontal direction, and adjacent plunger gates within the second electrode layer are also arranged staggered relative to each other in the horizontal direction. This staggered arrangement increases the effective separation distance between adjacent homogeneous gates within the same electrode layer and reduces mutual capacitance, thereby structurally minimizing horizontal electrostatic crosstalk within the layer. As a result, the tunnel coupling independent control precision of the barrier gates and the quantum dot potential independent control precision of the plunger gates are improved, thereby improving single-qubit and 2-qubit gate fidelity and extending the coherence time.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-14","filing_date":"2026-04-25","priority_date":"","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/40","G","G06","G06N","G06N10/00","G06N10/20","B","B82","B82Y","B82Y10/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260068731A/en"},{"publication_number":"KR20260068729A","title":"Neural processing device and Method for transmitting data thereof","abstract":"본 발명은 뉴럴 프로세싱 장치 및 그의 데이터 전송 방법을 개시한다. 상기 뉴럴 프로세싱 장치는, 제1 뉴럴 프로세서를 포함하는 적어도 하나의 뉴럴 프로세서, 상기 적어도 하나의 뉴럴 프로세서가 서로 공유하는 공유 메모리 및 상기 적어도 하나의 뉴럴 프로세서 및 상기 공유 메모리 사이의 데이터를 전송하는 글로벌 인터커넥션을 포함하고, 상기 제1 뉴럴 프로세서는, 제1 리드 리퀘스트를 생성하고, 제1 리퀘스트 ID를 가지는 제1 뉴럴 코어와, 상기 제1 리드 리퀘스트를 수신하고, 상기 제1 리드 리퀘스트에 대한 제2 리드 리퀘스트를 전송하는 로컬 인터커넥션과, 제2 리드 리퀘스트를 수신하고 상기 제2 리드 리퀘스트에 대한 리드 데이터를 수신하고, 상기 리드 데이터를 상기 적어도 하나의 뉴럴 코어에 전달하는 뉴럴 프로세서 캐시를 포함한다. The present invention discloses a neural processing device and a method for transmitting data therefrom. The neural processing device comprises at least one neural processor including a first neural processor, a shared memory shared by the at least one neural processor, and a global interconnection for transmitting data between the at least one neural processor and the shared memory. The first neural processor comprises a first neural core having a first request ID and generating a first read request, a local interconnection for receiving the first read request and transmitting a second read request for the first read request, and a neural processor cache for receiving the second read request, receiving read data for the second read request, and transmitting the read data to the at least one neural core.","assignee":"리벨리온 주식회사","inventors":["최성필","윤재성"],"publication_date":"2026-05-14","filing_date":"2026-04-24","priority_date":"","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06F","G06F12/00","G06F12/02","G06F12/08","G06F12/0802","G06F12/0893","G06F12/0895","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260068729A/en"},{"publication_number":"KR20260068044A","title":"Active Disturbance Security Apparatus Based on Computation Cycle Synchronization to Counter Electromagnetic Side-Channel Attacks on AI Computing Processors","abstract":"본 발명은 인공지능 연산 프로세서(10)의 전자기 부채널 공격에 대응하기 위한 능동 교란 보안 장치를 개시한다. 상기 장치는 프로세서(10)의 실시간 전력 소모 및 연산 주기를 모니터링하는 신호 감지부(100); 상기 연산 주기에 동기화되어 프로세서(10) 방출 전자기파와 동일한 주파수 대역의 교란 신호를 생성하는 신호 발생부(200); 및 프로세서(10)에 인접 배치되어 교란 신호를 외부로 방사하여 프로세서(10)의 고유 연산 파형을 은닉하는 교란 안테나부(300)를 포함한다. 연산 주기 동기화 방식은 종래 랜덤 노이즈 방식 대비 앙상블 평균 공격을 원천 차단하며, 추론/학습 모드 자동 판별, 폐루프 적응 제어 및 다층적 방어 체계를 통해 AI 프로세서 특화 보안 효과를 달성한다. The present invention discloses an active disturbance security device for responding to electromagnetic side-channel attacks on an artificial intelligence computation processor (10). The device includes: a signal detection unit (100) that monitors the real-time power consumption and computation cycle of the processor (10); a signal generation unit (200) that is synchronized with the computation cycle and generates a disturbance signal in the same frequency band as the electromagnetic waves emitted by the processor (10); and a disturbance antenna unit (300) that is positioned adjacent to the processor (10) and radiates the disturbance signal to the outside to conceal the unique computation waveform of the processor (10). The computation cycle synchronization method fundamentally blocks ensemble average attacks compared to the conventional random noise method and achieves AI processor-specific security effects through automatic determination of inference/learning mode, closed-loop adaptive control, and a multi-layered defense system.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-13","filing_date":"2026-04-26","priority_date":"","cpc_codes":["G","G06","G06F","G06F21/00","G06F21/50","G06F21/55","G06F21/556","G","G06","G06F","G06F21/00","G06F21/70","G06F21/71","G06F21/75","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260068044A/en"},{"publication_number":"KR20260068039A","title":"Heterogeneous Hybrid Computing System featuring Quantum Advantage Prediction-based Dynamic Task Classification, Local Closed-loop Error Correction, and GPU Digital Twins","abstract":"본 발명은 양자-고전 하이브리드 알고리즘을 실행하는 이종 하이브리드컴퓨팅 시스템(100)에 관한 것이다. 호스트 CPU(110)는 양자 이점 예측 모듈(111) 및 동적 분류기(112)를 통해 연산 태스크를 고전/양자로 능동적으로 분류하여 GPU(120) 및 QPU(130)에 각각 스케줄링한다. QPU(130) 내 로컬 제어기(131)는 호스트 CPU(110) 및GPU(120)의 개입 없이 독자적으로 오류 정정 피드백 루프를 수행하여 왕복 지연을 최소화한다. 반도체 스핀 큐비트 어레이(132)와 아날로그 CIM 기반NPU(133)가 TSV(134)를 통해 3D 적층되어, ADC 변환 없이 아날로그 측정 신호를 직접 처리하며 리저버 컴퓨팅 레이어(142)를 통한 디코히런스 선제 예측 보정을 수행한다. GPU(120)는QPU(130)의 디지털 트윈 노이즈 모델을 유지하여 양자 회로를 사전 최적화하고, GPU(120)와 QPU(130) 로컬 제어기(131) 사이의 저지연 직결 채널(123)이 변분 파라미터를 CPU(110) 우회하여 직접 전달한다. 확장 실시예로서 양자 노이즈 핑거프린트(150) 기반 하드웨어 인증, 큐비트 위상 재라우팅, qRAM 통합 메모리 계층, 연합 양자 학습 및 이중 모드 큐비트 감지-연산 직결 파이프라인(147)이 포함된다. The present invention relates to a heterogeneous hybrid computing system (100) that executes a quantum-classical hybrid algorithm. A host CPU (110) actively classifies computational tasks into classical and quantum through a quantum advantage prediction module (111) and a dynamic classifier (112) and schedules them to a GPU (120) and a QPU (130), respectively. A local controller (131) within the QPU (130) performs an error correction feedback loop independently without intervention from the host CPU (110) and the GPU (120) to minimize round-trip delay. A semiconductor spin qubit array (132) and an analog CIM-based NPU (133) are 3D stacked via a TSV (134) to directly process analog measurement signals without ADC conversion and perform decoherence preemptive prediction correction through a reservoir computing layer (142). The GPU (120) maintains the digital twin noise model of the QPU (130) to pre-optimize the quantum circuit, and a low-latency direct connection channel (123) between the GPU (120) and the QPU (130) local controller (131) directly transmits variational parameters by bypassing the CPU (110). As an extended embodiment, a quantum noise fingerprint (150)-based hardware authentication, qubit phase rerouting, qRAM integrated memory hierarchy, federated quantum learning, and a dual-mode qubit sense-computation direct connection pipeline (147) are included.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-13","filing_date":"2026-04-26","priority_date":"","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/70","G","G06","G06N","G06N10/00","G06N10/40","G","G06","G06N","G06N10/00","G06N10/60","B","B82","B82Y","B82Y10/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260068039A/en"},{"publication_number":"KR20260068040A","title":"Heterogeneous Hybrid Computing System with a CXL-based Unified Shared Memory Pool","abstract":"본 발명은 CXL(Compute Express Link) 프로토콜 기반의 인터커넥트 버스를 통해 CPU, GPU 및 QPU가 단일화된 공유 메모리 풀을 공유하는 이종 하이브리드 컴퓨팅 시스템을 개시한다. QPU는 CXL 엔드포인트로서 기능하는 양자 인터페이스 컨트롤러를 포함하며, 전용 양자 제어 컨트롤러의 개재 없이 CXL.mem 프로토콜을 통해 공유 메모리 풀의 지시서 스테이징 영역으로부터 양자 작업 디스크립터를 직접 리드하여 실행한다. 공유 메모리 풀은 CXL 스위치의 접근 제어 테이블에 의해 지시서 스테이징 영역, 결과 반환 영역 및 고전 전처리 영역으로 논리적으로 분할되며, GPU의 전처리 결과는 어드레스 포인터를 통한 제로카피 경로로 QPU에 직접 전달된다. 또한, 큐비트 코히런스 잔여 시간 정보에 연동한 양자 작업 디스크립터 선인출 메커니즘 및 양자 연산 부하 비율에 따른 메모리 풀 동적 재할당 기능을 포함한다. The present invention discloses a heterogeneous hybrid computing system in which a CPU, GPU, and QPU share a unified shared memory pool via an interconnect bus based on the Compute Express Link (CXL) protocol. The QPU includes a quantum interface controller that functions as a CXL endpoint and executes quantum work descriptors directly by reading them from the instruction staging area of the shared memory pool via the CXL.mem protocol without the intermediation of a dedicated quantum control controller. The shared memory pool is logically divided into an instruction staging area, a result return area, and a quantum preprocessing area by an access control table of a CXL switch, and the preprocessing results of the GPU are directly transmitted to the QPU via a zero-copy path through an address pointer. Additionally, it includes a quantum work descriptor pre-fetch mechanism linked to qubit coherence remaining time information and a memory pool dynamic reallocation function based on the quantum computation load ratio.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-13","filing_date":"2026-04-26","priority_date":"","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/40","G","G06","G06N","G06N10/00","G06N10/60","B","B82","B82Y","B82Y10/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260068040A/en"},{"publication_number":"KR20260068041A","title":"Quantum Computing Device Featuring Spatially Separated Nuclear Spin-Based Quantum Memory via Deterministic Single-Ion Implantation","abstract":"본 발명은 동위원소 정제된 28 Si 반도체 기판 내에 형성된 양자점의 단일 전자 스핀을 논리 큐비트로 활용하고, 결정론적 단일 이온 주입 공정에 의해 양자 우물 영역의 외부 경계면으로부터 10 nm 이상 20 nm 이하의 이격 거리에 배치된 단 하나의 ³¹P 불순물 원자의 원자핵 스핀을 메모리 큐비트로 활용하는 공간 분리형 이중 계층 양자 메모리 아키텍처를 갖는 양자 컴퓨팅 장치에 관한 것이다. 양자 연산 모드에서 논리 큐비트와 메모리 큐비트는 전기적·자기적으로 완전히 디커플링되어 상호 결어긋남이 억제되며, 저장 모드에서는 ESR 펄스와 NMR 펄스의 이중 펄스 시퀀스를 통한 공명 SWAP 게이트에 의해 전자의 양자 상태가 원자핵으로 고속·고충실도로 전이되어 수 초 이상 보존된다. 3차원 적층된 아날로그 CIM NPU가 ADC 지연 없이 즉각적으로 SWAP 트리거를 생성하여 연산 결과의 붕괴를 방지하며, 2차원 양자점 어레이와 크로스바 어드레싱 구조에 의해 대규모 양자 랜덤 접근 메모리(QRAM)로 확장 가능하다. The present invention relates to a quantum computing device having a spatially separated dual-layer quantum memory architecture that utilizes a single electron spin of a quantum dot formed within an isotope-purified 28 Si semiconductor substrate as a logic qubit, and utilizes a single nucleus spin of a single ³¹P impurity atom disposed at a distance of 10 nm to 20 nm from the outer boundary of the quantum well region by a deterministic single-ion implantation process as a memory qubit. In quantum operation mode, the logic qubit and the memory qubit are completely electrically and magnetically decoupled to suppress mutual decoupling, and in storage mode, the quantum state of the electron is transferred to the nucleus at high speed and high fidelity by a resonance SWAP gate through a dual pulse sequence of ESR pulses and NMR pulses, and is preserved for several seconds or more. A 3D stacked analog CIM NPU generates a SWAP trigger immediately without ADC delay to prevent the collapse of the operation result, and can be expanded into a large-scale quantum random access memory (QRAM) by means of a 2D quantum dot array and a crossbar addressing structure.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-13","filing_date":"2026-04-26","priority_date":"","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/40","G","G06","G06N","G06N10/00","G06N10/70","B","B82","B82Y","B82Y10/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260068041A/en"},{"publication_number":"KR20260068037A","title":"Non-contact Spin-state Readout Apparatus Utilizing Changes in Magnetic Susceptibility of a Single Electron within a Semiconductor Quantum Dot","abstract":"본 발명은 반도체 양자점(100) 내 단일 전자(140)의 스핀 상태를 자기 채널만으로비접촉 판독하는 양자 컴퓨팅 장치에 관한 것이다. 양자점 트랩부(100)는 복수의 게이트 전극(120)에 의한 전위 우물(130)로 단일 전자(140)를 고립시키고, 미세 전자석부(200)의 코일 구조(210)가 LC 탱크 회로(310)의 인덕터 소자를 겸하여 정자기장 인가와 자기 감수율 감지를 이중으로 수행한다. 단일 전자(140)의 스핀 상태 전환에 따른 자기 감수율 변화(Δχ)가 코일 구조(210)의 인덕턴스(L)를 변조하고, 이 인덕턴스 변화(ΔL)가 자기 임피던스 판독부(300)의 임피던스 변화로 나타난다. 측정 회로(320)의 IQ 복조기(321)가 위상 변화량으로 스핀 상태를 판별한다. 초전도 운동 인덕턴스 증폭, 이중 코일 분리, 자속 잠금 루프(330), 파라메트릭 증폭기(340), 차동 이중 양자점, 주파수 다중화 어레이(600), 라머 동기 스트로보스코픽 감지, 퀴리 법칙 온도 보정, DNP 협력 증폭, 텐서 비대칭 저자기장 판독, 및 역방향 작용 조형 자기 증폭을 통해 감도·충실도·확장성을 종합적으로향상시킨다. The present invention relates to a quantum computing device that reads the spin state of a single electron (140) within a semiconductor quantum dot (100) non-contactually using only a magnetic channel. The quantum dot trap section (100) isolates the single electron (140) into a potential well (130) formed by a plurality of gate electrodes (120), and the coil structure (210) of the microelectromagnet section (200) doubles as an inductor element of the LC tank circuit (310) to perform the dual application of a static magnetic field and detection of magnetic susceptibility. The change in magnetic susceptibility (Δχ) due to the spin state transition of the single electron (140) modulates the inductance (L) of the coil structure (210), and this change in inductance (ΔL) appears as a change in impedance of the magnetic impedance reading section (300). The IQ demodulator (321) of the measurement circuit (320) determines the spin state based on the amount of phase change. Sensitivity, fidelity, and scalability are comprehensively improved through superconducting kinetic inductance amplification, dual coil separation, flux-locked loop (330), parametric amplifier (340), differential dual quantum dot, frequency multiplexing array (600), Larmor synchronous stroboscopic detection, Curie law temperature correction, DNP cooperative amplification, tensor asymmetric low magnetic field reading, and reverse action shaping magnetic amplification.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-13","filing_date":"2026-04-25","priority_date":"","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/40","B","B82","B82Y","B82Y10/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260068037A/en"},{"publication_number":"KR20260068036A","title":"Apparatus and Method for Non-mediated Non-destructive Quantum State Detection of Ion Qubits via Motional Modes","abstract":"본 발명은 이온 기반 큐비트의 양자 상태를 비파괴적으로 검출하는 장치 및 방법에 관한 것이다. 본 발명의 장치는 양자 중첩 또는 양자 얽힘 상태를 유지하도록 전자기적 트랩 내에 부양된 이온 기반 큐비트부, 큐비트의 전하 분포에 대응하는 이미지 전하가 표면에 유도되는 검출 전극, 큐비트의 양자 상태에 따른 미세 전자기 간섭 신호를 감지하는 고감도 센서부, 및 감지된 신호를 분석하여 큐비트의 양자 상태를 판별하는 제어부를 포함한다. 상기 이온 기반 큐비트에 측정용 광자 에너지를 인가하지 않고 전자기적 유도 결합만을 이용하여 큐비트 양자 상태를 판독함으로써 파동함수 붕괴를 최소화하고, 이온의 운동 모드를 매개로 하지 않으므로 운동 모드가 후속 양자 게이트 연산에 온전히 보존된다. 이로써 레이저 형광 검출 방식 대비 결맞음 보존, 운동 모드 보존, 중간 측정 가능, 레이저 자유 집적화라는 현저한 효과가 달성된다. The present invention relates to an apparatus and method for non-destructively detecting the quantum state of an ion-based qubit. The apparatus of the present invention comprises an ion-based qubit portion levitating within an electromagnetic trap to maintain a quantum superposition or quantum entanglement state, a detection electrode on which an image charge corresponding to the charge distribution of the qubit is induced on its surface, a high-sensitivity sensor portion detecting a micro-electromagnetic interference signal according to the quantum state of the qubit, and a control portion analyzing the detected signal to determine the quantum state of the qubit. By reading the qubit quantum state using only electromagnetic inductive coupling without applying measurement photon energy to the ion-based qubit, wave function decay is minimized, and since the motion mode of the ion is not mediated, the motion mode is fully preserved in subsequent quantum gate operations. This achieves significant effects compared to laser fluorescence detection methods, such as coherence preservation, motion mode preservation, intermediate measurement capability, and laser-free integration.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-13","filing_date":"2026-04-25","priority_date":"","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/40","B","B82","B82Y","B82Y10/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260068036A/en"},{"publication_number":"KR20260067364A","title":"Analog feedback control-based semiconductor quantum computing device","abstract":"본 발명은 반도체 스핀 큐비트 기반의 양자 컴퓨팅 장치에 관한 것으로, 다층 게이트 뱅크 구조의 반도체 양자점 어레이(100), HEMT 기반의 저온 증폭부(200), 아날로그 CIM 기반의 NPU(300), 및 제어부(400)를 포함한다. 저온 증폭부(200)는 큐비트의 아날로그 측정 신호를 레벨 시프터(220)를 통해 NPU(300)의 메모리 셀 어레이(310)에 ADC 없이 직접 인가하고, NPU(300)는 옴의 법칙 기반 전류 합산으로 아날로그 MAC 연산을 수행하여 보정 값을 산출하며, 제어부(400)의 DAC(410) 및 게이트 드라이버(420)를 통해 게이트 전압을 실시간으로 보정하는 완전 아날로그 피드백 루프를 형성한다. 이를 통해 피드백 지연을 수십 ns 수준으로 단축하여 큐비트 코히어런스 시간 내에 오류 정정 루프를 완결하며, 3D 적층 패키징 및 TSV(500) 기반 직결 구조로 배선 지연을 최소화한다. The present invention relates to a semiconductor spin qubit-based quantum computing device comprising a semiconductor quantum dot array (100) having a multilayer gate bank structure, a HEMT-based low-temperature amplifier (200), an analog CIM-based NPU (300), and a control unit (400). The low-temperature amplifier (200) directly applies an analog measurement signal of a qubit to a memory cell array (310) of the NPU (300) without an ADC through a level shifter (220), and the NPU (300) calculates a correction value by performing an analog MAC operation based on Ohm's law current summation, and forms a fully analog feedback loop that corrects the gate voltage in real time through a DAC (410) and a gate driver (420) of the control unit (400). Through this, the feedback delay is reduced to the level of tens of nanoseconds to complete the error correction loop within the qubit coherence time, and wiring delay is minimized through a direct connection structure based on 3D stacked packaging and TSV (500).","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-12","filing_date":"2026-04-25","priority_date":"","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/40","G","G06","G06N","G06N10/00","G06N10/20","G","G06","G06N","G06N10/00","G06N10/60","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","B","B82","B82Y","B82Y10/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260067364A/en"},{"publication_number":"KR20260066016A","title":"Privacy-Preserving Expert Module Anonymized Extraction System","abstract":"본 발명은 기업 내부망에서 민감 데이터로 학습된 전문가 모듈의 가중치로부터 민감 정보를 제거하고 외부 마켓플레이스에 유통 가능한 형태로 변환하는 프라이버시 보존형 전문가 모듈 익명화 추출 시스템에 관한 것이다. 본 발명은 원본 학습 데이터 및 모든 중간 산출물을 내부망에 보유하는 온프레미스 학습 엔진; 그래디언트 기여도 분석·영향 함수·카나리 탐지·LRP 등 다중 수단으로 고위험 파라미터를 식별하는 민감 정보 인코딩 탐지기; 고위험 파라미터에만 선별적으로 DP 노이즈·평균값 대체·SVD 근사·경량 재훈련·참조 모델 대체를 적용하는 가중치 공간 프라이버시 필터; 멤버십 추론·모델 역전·그래디언트 역전·속성 추론 공격 시뮬레이션으로 수치 검증하고 규제 등급별 프라이버시 보증 증명서를 발급하는 익명화 검증기; 및 자기 증명형 판매 단위를 생성하는 시장 출시 변환기를 포함하며, 제조·금융·의료 기업들이 민감 노하우를 AI 가중치 형태로 안전하게 상업화할 수 있는 기술 인프라를 제공한다. The present invention relates to a privacy-preserving expert module anonymization extraction system that removes sensitive information from the weights of an expert module trained with sensitive data in an internal corporate network and converts it into a form that can be distributed in an external marketplace. The present invention includes an on-premise learning engine that holds original training data and all intermediate outputs in an internal network; a sensitive information encoding detector that identifies high-risk parameters using multiple means such as gradient contribution analysis, influence function, canary detection, and LRP; a weight space privacy filter that selectively applies DP noise, mean imputation, SVD approximation, lightweight retraining, and reference model imputation only to high-risk parameters; an anonymization validator that performs numerical verification through membership inference, model inversion, gradient inversion, and attribute inference attack simulations and issues privacy assurance certificates by regulatory grade; and a market-to-market converter that generates self-certified sales units, thereby providing a technical infrastructure that enables manufacturing, financial, and medical companies to safely commercialize sensitive know-how in the form of AI weights.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-12","filing_date":"2026-04-20","priority_date":"","cpc_codes":["G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6245","G06F21/6254","G","G06","G06F","G06F21/00","G06F21/50","G06F21/57","G06F21/577","G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260066016A/en"},{"publication_number":"KR20260065601A","title":"AI computing apparatus for side-channel attack defense based on random sampling and dummy bursts","abstract":"본 발명은 인공지능 모델의 가중치 연산 시 발생하는 부채널 신호를 무작위 샘플링 기반 더미 버스트 신호를 이용하여 은닉하는 인공지능 연산 장치에 관한 것으로, 인공지능 모델의 실제 가중치 연산 시퀀스를 모니터링하고 교란 신호 주입 지점을 결정하는 보안 스케줄러(110), 비정기적이고 무작위인 시점에 실제 연산 파형과 유사한 특성의 더미 버스트 신호를 생성하는 노이즈 발생부(120), 및 더미 버스트 신호를 실제 연산 신호에 중첩시켜 외부에서 측정되는 전자기파 신호의 통계적 유의성을 왜곡하는 신호 합성부(130)를 포함하며, 더미 버스트 신호는 전체 연산 시퀀스의 일부 샘플링된 구간에서만 간헐적으로 발생하여 연산 오버헤드를 5% 미만으로 유지하면서도 외부 공격자의 평균화 기반 통계 분석을 원천적으로 차단하는 효과를 제공한다. The present invention relates to an artificial intelligence computing device that conceals side-channel signals generated during weight calculation of an artificial intelligence model using random sampling-based dummy burst signals, comprising a security scheduler (110) that monitors the actual weight calculation sequence of the artificial intelligence model and determines a point for injecting a disturbance signal, a noise generator (120) that generates a dummy burst signal with characteristics similar to the actual calculation waveform at irregular and random times, and a signal synthesis unit (130) that superimposes the dummy burst signal onto the actual calculation signal to distort the statistical significance of an electromagnetic wave signal measured from the outside, wherein the dummy burst signal is generated intermittently only in a part of a sampled section of the entire calculation sequence, thereby maintaining the calculation overhead at less than 5% while providing the effect of fundamentally blocking statistical analysis based on averaging by external attackers.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-08","filing_date":"2026-04-21","priority_date":"","cpc_codes":["G","G06","G06F","G06F21/00","G06F21/50","G06F21/55","G06F21/556","G","G06","G06F","G06F21/00","G06F21/70","G06F21/71","G06F21/75","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260065601A/en"},{"publication_number":"KR20260065602A","title":"Apparatus and method for hiding side-channel signals of an AI computing device using resource-efficient low-precision pseudo-weight operations","abstract":"본 발명은 인공지능 모델의 가중치 연산 과정에서 발생하는 전자기파 방사 기반 부채널 신호를 은닉하는 장치 및 방법에 관한 것으로서, 프로세서의 자원 점유율 및 유휴 상태를 실시간으로 모니터링하는 자원 감시부(110), 실제 가중치와 통계적 파형 특성은 유사하되 연산 복잡도가 낮은 저정밀도 의사 가중치 데이터를 생성하는 의사 데이터 생성부(120), 및 식별된 유휴 자원 범위 내에서 실제 가중치 연산 시퀀스 사이에 저정밀도 의사 연산을 동적으로 삽입하는 연산 제어부(130)를 포함한다. 의사 연산은 4비트 이하 저정밀도 또는 이진 연산으로 수행되어 전력 소모를 70% 이상 절감하고, 파이프라인 버블 및 메모리 대기 시간의 유휴 사이클을 활용하여 추론 지연을 발생시키지 않으며, 이미 로드된 실제 가중치의 비트 교환 방식으로 의사 가중치를 재사용하여 메모리 대역폭을 절약한다. 고부하 상태에서는 소프트맥스 레이어 또는 활성화 함수 구간으로 보안 범위를 동적으로 축소하며, 저전력 하위 연산 유닛(142)만 선택적으로 활성화하고 고성능 연산 유닛(141)의 클럭을 차단한다. The present invention relates to an apparatus and method for concealing electromagnetic radiation-based side-channel signals generated during the weight calculation process of an artificial intelligence model. It comprises a resource monitoring unit (110) that monitors the resource occupancy and idle state of a processor in real time, a pseudo-data generation unit (120) that generates low-precision pseudo-weight data with low computational complexity that is similar to actual weights and statistical waveform characteristics, and a calculation control unit (130) that dynamically inserts low-precision pseudo-calculations between actual weight calculation sequences within an identified idle resource range. The pseudo-calculations are performed as low-precision or binary operations of 4 bits or less, thereby reducing power consumption by more than 70%, utilizing idle cycles of pipeline bubbles and memory waiting times to avoid inference delays, and saving memory bandwidth by reusing pseudo-weights through a bit exchange method of already loaded actual weights. In a high-load state, the security range is dynamically reduced to a softmax layer or an activation function section, and only the low-power lower-computation unit (142) is selectively activated while the clock of the high-performance computation unit (141) is blocked.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-08","filing_date":"2026-04-21","priority_date":"","cpc_codes":["G","G06","G06F","G06F21/00","G06F21/50","G06F21/55","G06F21/556","G","G06","G06F","G06F21/00","G06F21/70","G06F21/71","G06F21/75","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260065602A/en"},{"publication_number":"KR20260065598A","title":"Method and system for tokenizing artificial intelligence model weights linked to real-world assets","abstract":"본 발명은 학습이 완료된 인공 지능 모델의 가중치 데이터를 실물 자산 연동 토큰(RWA 토큰)으로 변환하는 방법 및 시스템에 관한 것이다. 본 발명에 따르면, 가중치 데이터의 암호화 해시값을 생성하고, 암호화된 가중치를 오프체인 보안 저장소에 저장하며, 가중치 메타데이터와 RWA 토큰을 블록체인에 기록하고, 토큰 보유 증명에 따라 신뢰 실행 환경(TEE) 내에서만 가중치에 대한 접근을 허용한다. 특히, 파인튜닝 등에 의해 지속적으로 업데이트되는 동적 가중치의 특성을 수용하기 위하여 스냅샷 버전 토큰화, 차분 데이터 분리 토큰화, 거버넌스 승인 기반 업데이트, 및 살아있는 자산 토큰 구조를 제공한다. 또한, 추론 실행 권한, 파인튜닝 권한, 거버넌스 의결권을 유형별로 분리 발행하고, 스마트컨트랙트를 통한 자동 수익 분배 및 파인튜닝 로열티 자동 지급을 실현한다. The present invention relates to a method and system for converting weight data of a trained artificial intelligence model into real-world asset-linked tokens (RWA tokens). According to the present invention, an encrypted hash value of the weight data is generated, the encrypted weights are stored in an off-chain secure storage, weight metadata and RWA tokens are recorded on a blockchain, and access to the weights is allowed only within a Trusted Execution Environment (TEE) based on proof of token ownership. In particular, to accommodate the characteristics of dynamic weights that are continuously updated by fine-tuning, the invention provides snapshot version tokenization, differential data separation tokenization, governance approval-based updates, and a living asset token structure. Furthermore, inference execution rights, fine-tuning rights, and governance voting rights are issued separately by type, and automatic profit distribution and automatic payment of fine-tuning royalties are realized through smart contracts.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-08","filing_date":"2026-04-21","priority_date":"","cpc_codes":["H","H04","H04L","H04L9/00","H04L9/32","H04L9/321","H04L9/3213","G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G","G06","G06F","G06F21/00","G06F21/60","G06F21/64","G","G06","G06N","G06N20/00","H","H04","H04L","H04L9/00","H04L9/32","H04L9/3236","H","H04","H04L","H04L9/00","H04L9/50"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260065598A/en"},{"publication_number":"KR20260065600A","title":"Apparatus and method for defending AI models against side-channel attacks via low-power dummy signal synthesis and selective interval protection","abstract":"본 발명은 인공지능 모델의 가중치를 부채널 공격으로부터 보호하는 인공지능 연산 장치 및 방법에 관한 것이다. 본 발명에 따른 인공지능 연산 장치는 인공지능 연산의 전체 시퀀스 중 부채널 신호 유출 위험이 높은 특정 연산 구간을 선택적으로 식별하는 보안 제어부, 상기 구간 내에서 실제 가중치 연산과 전기적 특성이 매칭되는 저전력 더미 신호를 생성하는 노이즈 발생부, 및 실제 연산 결과에는 영향을 주지 않으면서 외부로 방출되는 전자기파 지문에만 상기 더미 신호를 중첩시키는 신호 합성부를 포함한다. 더미 신호의 연산량은 전체 실제 연산량의 미리 설정된 임계치 이내로 제한되어 보안 오버헤드를 최소화하며, 입력층 및 출력층에 보안을 집중하는 레이어별 차등 보호, 더미 로드 회로 기반의 비연산 더미 신호 생성, 비트 반전 방식의 초경량 더미 신호 생성, 유휴 자원 기반의 적응형 보안 제어 및 타이밍 지터 삽입 기능을 통해 배터리 구동 엣지 디바이스에서도 실용적으로 부채널 공격을 방어할 수 있다. The present invention relates to an artificial intelligence computing device and method for protecting the weights of an artificial intelligence model from side-channel attacks. The artificial intelligence computing device according to the present invention includes a security control unit that selectively identifies a specific computation section with a high risk of side-channel signal leakage within the entire sequence of artificial intelligence computations, a noise generation unit that generates a low-power dummy signal whose electrical characteristics match the actual weight computation within said section, and a signal synthesis unit that superimposes the dummy signal only on the electromagnetic fingerprint emitted externally without affecting the actual computation result. The computational amount of the dummy signal is limited to within a preset threshold of the total actual computation amount to minimize security overhead, and through layer-specific differential protection concentrating security on the input and output layers, non-computational dummy signal generation based on a dummy load circuit, ultra-lightweight dummy signal generation using a bit inversion method, adaptive security control based on idle resources, and timing jitter insertion functions, it is possible to practically defend against side-channel attacks even on battery-powered edge devices.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-08","filing_date":"2026-04-21","priority_date":"","cpc_codes":["G","G06","G06F","G06F21/00","G06F21/50","G06F21/55","G06F21/556","G","G06","G06F","G06F21/00","G06F21/70","G06F21/71","G06F21/75","G06F21/755","G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260065600A/en"},{"publication_number":"KR20260064680A","title":"Query processing method and system","abstract":"본 발명은 사용자의 질의 의도에 따른 응답을 출력하기 위한 질의 처리 방법 및 시스템에 관한 것이다. 본 발명에 따른 질의 처리 방법은, 사용자로부터 사용자 질의를 수신하는 단계, 적어도 하나 이상의 문장을 입력으로 하여 정규화 된 하나의 문장 또는 단어를 생성하도록 학습된 질의 정규화 모델을 이용하여, 상기 사용자 질의에 상응하는 정규화된 질의를 생성하는 단계, 상기 정규화된 질의에 대응하는 응답을 생성하는 단계 및 상기 응답을 상기 사용자에게 제공하는 단계를 포함할 수 있다. The present invention relates to a query processing method and system for outputting a response according to a user's query intent. A query processing method according to the present invention may include the steps of receiving a user query from a user, generating a normalized query corresponding to the user query using a query normalization model trained to generate a normalized sentence or word with at least one sentence as input, generating a response corresponding to the normalized query, and providing the response to the user.","assignee":"네이버 주식회사","inventors":["이유영","김혜영","김선라","인수교","문기윤","김경덕","서수빈","남경민","김현욱"],"publication_date":"2026-05-07","filing_date":"2026-04-27","priority_date":"","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/3332","G06F16/3334","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F16/00","G06F16/30","G06F16/34","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481","G06F3/04817","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G06F40/211","G","G06","G06F","G06F40/00","G06F40/30","G06F40/35","G","G06","G06N","G06N20/00","G","G10","G10L","G10L15/00","G10L15/26"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260064680A/en"},{"publication_number":"AU2026202952A1","title":"Metadata tag auto-application to posted entries","abstract":"A system and a method are disclosed for receiving, from a source of a plurality of candidate sources, a payload comprising content and metadata. The system selects a destination to which to route the payload based on the source and the content, and generates an entry at the destination based on the content. The system inputs the metadata into a classification engine, and receives, as output from the classification engine, one or more classifications for the payload. The system applies a metadata tag to the entry, the metadata tag indicating the one or more classifications. The system receives a search request from a client device specifying at least one of the one or more classifications, and, in response to receiving the search request, provides the entry to the client device based on a matching classification.","assignee":"Tekion Corp","inventors":["Satyavrat Mudgil","Anant Sitaram"],"publication_date":"2026-05-07","filing_date":"2026-04-20","priority_date":"2021-02-12","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G06F16/353","G","G06","G06F","G06F16/00","G06F16/20","G06F16/28","G06F16/284","G06F16/285","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G","G06","G06F","G06F16/00","G06F16/30","G06F16/38","G","G06","G06N","G06N20/00","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N5/00","G06N5/04"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202952A1/en"},{"publication_number":"AU2026202913A1","title":"Machine-learned validation framework and entity function management","abstract":"46 A system and a method are disclosed herein for machine-learned detection of outliers within payload requests. An entity management system uses machine learning to cluster data characterizing requests from entities to route payloads, and determines one or more data clusters that are outliers. The system receives a request to route a payload to a destination, and applies a supervised machine learning model to size and type information indicated by the payload. The supervised machine learning model applies a label to the payload data (e.g., indicating that the payload routing request is an outlier). This outlier detection may drive a validation process to address detected outliers. The system may receive an indication to perform a validation function and transmit the payload to a validation destination. The system may leverage payload data and feedback received from an entity to optimize machine learning techniques to the entity.","assignee":"Tekion Corp","inventors":["Satyavrat Mudgil","Anant Sitaram","Ved Surtani"],"publication_date":"2026-05-07","filing_date":"2026-04-17","priority_date":"2021-04-05","cpc_codes":["G","G06","G06N","G06N20/00"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202913A1/en"},{"publication_number":"AU2026202865A1","title":"Systems and methods for supply chain management including erp agnostic realtime data mesh with change data capture","abstract":"System and methods are provided for dynamically consolidating interaction points in a distribution ecosystem. The method involves integrating multiple touchpoints of communication between distributors, resellers, end-users, vendors, and suppliers into a unified interactive interface. This interface enables the management of end-to-end partner lifecycle, systematic data collection, analysis using advanced statistical algorithms, deployment of artificial intelligence and machine learning algorithms, and continuous updates based on user feedback. The system includes modules for communication integration, consolidation, lifecycle management, data collection, data analysis, and artificial intelligence. The disclosed method and system enhance supply chain operations, generate actionable insights, and provide personalized user experiences, ultimately driving business growth and efficiency. 20 26 20 28 65 16 A pr 2 02 6 A B S T R A C T 2 0 2 6 2 0 2 8 6 5 1 6 A p r 2 0 2 6","assignee":"Ingram Micro Inc","inventors":["Mukund Gopalan","Sanjib Sahoo"],"publication_date":"2026-05-07","filing_date":"2026-04-16","priority_date":"2023-07-10","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0605"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202865A1/en"},{"publication_number":"AU2026202818A1","title":"Methods and apparatus for assessing candidates for visual roles","abstract":"The techniques described herein relate to methods, apparatus, and computer readable media configured to receive a set of images associated with a candidate, wherein each image is a visual work created by the candidate, and process the set of images using one or more machine learning techniques, artificial intelligence techniques, or both, to add the set of images to a search 5 index.","assignee":"Aquent LLC","inventors":["Patrick Branigan","Brennan CARLSON","Kimberley CROSCUP","Shanthi GUDIGOPURAM","Maxfield Howes","Zachary HUNTER","Nadav LAPIDOT","Tim Mays","Lauren PEHNKE","Frank ROMEU","Daryl RUE","Matthew TONEY","Eric WITHERSPOON","Jeremy Wood","Yan Yan"],"publication_date":"2026-05-07","filing_date":"2026-04-15","priority_date":"2019-09-23","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/105","G06Q10/1053","G","G06","G06F","G06F16/00","G06F16/50","G06F16/51","G","G06","G06F","G06F16/00","G06F16/50","G06F16/53","G06F16/538","G","G06","G06F","G06F16/00","G06F16/50","G06F16/58","G06F16/583","G06F16/5846","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G","G06","G06N","G06N20/00","G","G06","G06T","G06T7/00","G06T7/10","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20021","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202818A1/en"},{"publication_number":"AU2026202843A1","title":"AI-assisted medical image interpretation and report generation","abstract":"Disclosed herein are systems, methods, and software for providing a platform for AI-assisted medical image interpretation and report generation. One or more subsystems allow for the capturing of user input such as eye gaze and dictation for automated generation of clinical findings. Additional features include quality metric tracking and feedback, and worklist management system and communications queueing.","assignee":"Sirona Medical Inc","inventors":["Cameron Andrews","Ankit Goyal","Mark D. LONGO","Vernon Marshall","Berk NORMAN","Kojo Worai OSEI","David Seungwon PAIK"],"publication_date":"2026-05-07","filing_date":"2026-04-15","priority_date":"2019-10-01","cpc_codes":["G","G16","G16H","G16H15/00","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G06F18/2148","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/40","G06F18/41","G","G06","G06F","G06F3/00","G06F3/01","G06F3/011","G06F3/013","G","G06","G06F","G06F3/00","G06F3/16","G06F3/167","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/091","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G06","G06V","G06V10/00","G06V10/94","G06V10/95","G","G10","G10L","G10L15/00","G10L15/22","G","G16","G16H","G16H30/00","G16H30/40","G","G06","G06T","G06T2200/00","G06T2200/24","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10056","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10068","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10072","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10116"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202843A1/en"},{"publication_number":"AU2026202827A1","title":"Reducing domain shift in neural motion controllers","abstract":"1006522704 One embodiment of the present invention sets forth a technique for training a neural motion controller. The technique includes determining a first set of features associated with a first control signal for a virtual character. The technique also includes matching the first set of features to a first sequence of motions included in a plurality of sequences of motions. The technique further includes training the neural motion controller based on one or more motions included in the first sequence of motions and the first control signal.","assignee":"Eidgenoessische Technische Hochschule Zurich ETHZ; Disney Enterprises Inc","inventors":["Dhruv Agrawal","Dominik Tobias Borer","Jakob Joachim Buhmann","Martin Guay","Mattia Gustavo Bruno Paolo Ryffel","Robert Walker Sumner"],"publication_date":"2026-05-07","filing_date":"2026-04-15","priority_date":"2023-02-21","cpc_codes":["G","G06","G06T","G06T13/00","G06T13/20","G06T13/40","G","G06","G06T","G06T13/00","G","G06","G06N","G06N3/00","G06N3/02","G","G06","G06T","G06T7/00","G06T7/20","G06T7/246","G06T7/248","G","G06","G06T","G06T7/00","G06T7/70","G06T7/73","G06T7/74","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10016","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30196"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202827A1/en"},{"publication_number":"KR102961138B1","title":"Real-time video processing system and method for an edge device","abstract":"엣지 디바이스에서의 실시간 영상 처리 시스템 및 방법이 제공된다. 본 발명의 실시예들에 따르면, 영상 프레임이 메모리 상에서 이동하지 않고 물리주소를 기반으로 전달됨으로써 CPU에 의한 메모리 복사 없이 영상 처리가 가능해지므로, 메모리 대역폭의 사용량이 감소하며 이에 따라 전체 처리 성능이 향상될 수 있다. 특히, 영상 프레임이 입력 단계부터 출력 단계까지 Zero-copy 방식으로 처리됨에 따라 지연시간(latency)이 감소하고, 고해상도 영상에 대해서도 안정적인 실시간 처리가 가능해진다. 이 경우, CPU가 대용량 데이터 복사 작업에서 해방될 수 있어 후처리, 객체 트래킹 등의 고수준 연산에 CPU 자원을 집중 배분할 수 있게 되므로, 전체 시스템 처리 효율이 향상될 수 있다. A system and method for real-time image processing on an edge device are provided. According to embodiments of the present invention, image frames are transmitted based on physical addresses without moving in memory, thereby enabling image processing without memory copying by the CPU. Consequently, memory bandwidth usage is reduced, and overall processing performance can be improved. In particular, as image frames are processed in a zero-copy manner from the input stage to the output stage, latency is reduced, and stable real-time processing is possible even for high-resolution images. In this case, the CPU can be freed from large-scale data copying tasks, allowing CPU resources to be concentrated on high-level operations such as post-processing and object tracking, thereby improving the overall system processing efficiency.","assignee":"주식회사 티솔루션즈","inventors":["이명철"],"publication_date":"2026-05-06","filing_date":"2026-04-16","priority_date":"","cpc_codes":["G","G06","G06T","G06T1/00","G06T1/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G","G06","G06T","G06T1/00","G06T1/60"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102961138B1/en"},{"publication_number":"KR20260061126A","title":"Phononic Crystal-based n-bit Weight Neuromorphic Device and Phononic Neuromorphic Computing Array Using the Same","abstract":"본 발명은 포노닉 결정(Phononic Crystal) 구조의 변조부(120)에서 음향파의 위상차 및/또는 진폭을 n단계(n은 2 이상의 정수)로 조절함으로써 n가지 안정 상태를 구현하고, 상기 n가지 안정 상태를 각각 인공신경망의 n진 가중치에 대응시켜 log₂(n) 비트의 가중치를 저장하는 뉴로모픽 소자(100) 및 상기 소자를 크로스바 어레이(200) 형태로 집적하여 음향 신호와 포노닉 가중치의 상호작용에 의한 곱셈-누산 연산을 수행하는 포노닉 뉴로모픽 컴퓨팅 어레이(200)를 제공한다. n=3인 삼진 경우 BitNet b1.58 아키텍처와 직접 호환되며, 압전 소자(124), 열제어 소자(125), 결함(123) 고정, 자기탄성 효과 등 다양한 격자 상수(122) 제어 방식이 제공된다. The present invention provides a neuromorphic element (100) that implements n stable states by adjusting the phase difference and/or amplitude of an acoustic wave in n steps (n is an integer greater than or equal to 2) in a modulation unit (120) of a phononic crystal structure, and stores a log₂(n) bit weight corresponding to each of the n stable states and an n-ary weight of an artificial neural network, and a phononic neuromorphic computing array (200) that integrates the element in the form of a crossbar array (200) to perform a multiplication-accumulation operation based on the interaction between the acoustic signal and the phononic weight. In the case of n=3, it is directly compatible with the BitNet b1.58 architecture, and various lattice constant (122) control methods are provided, such as a piezoelectric element (124), a thermal control element (125), a defect (123) fixation, and a magneto-elastic effect.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-06","filing_date":"2026-04-15","priority_date":"","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G06N3/065","G","G10","G10K","G10K11/00","G10K11/36"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260061126A/en"},{"publication_number":"KR20260060269A","title":"Ddc controller device for building automation control having redundancy authority management function","abstract":"본 발명은 BAS/BEMS용 DDC 컨트롤러 장치(1000)에 관한 것으로, 메인 제어부(100), 엣지 AI 추론 엔진부(200), 멀티프로토콜 통합 통신 모듈(300), 유니버설 I/O 모듈(400), 에너지 미터링 인터페이스부(500) 및 이중화 권한 관리부(600)를 포함한다. 이중화 권한 관리부(600)는 Master 및 Slave 양측과 양방향 통신 경로를 형성하여 통신이상을 개별 감시하는 5개 하위 모듈을 구비하며, 확인 및 승인 처리부(620)가 AI 모델 상태와 에너지 데이터를 포함하는 확장 동기화 데이터셋의 무결성을 검증한 후 권한 토큰을 발행하여 전환함으로써, 무중단 운전과 AI 제어 연속성을 보장한다. The present invention relates to a DDC controller device (1000) for BAS/BEMS, comprising a main control unit (100), an edge AI inference engine unit (200), a multi-protocol integrated communication module (300), a universal I/O module (400), an energy metering interface unit (500), and a redundancy authority management unit (600). The redundancy authority management unit (600) is equipped with five sub-modules that individually monitor communication anomalies by forming a bidirectional communication path with both the Master and Slave sides, and the verification and approval processing unit (620) ensures uninterrupted operation and AI control continuity by issuing an authority token and switching after verifying the integrity of an extended synchronization dataset containing AI model status and energy data.","assignee":"유한회사 한국기계설비기술","inventors":["정진홍"],"publication_date":"2026-05-04","filing_date":"2026-04-14","priority_date":"","cpc_codes":["H","H04","H04L","H04L41/00","H04L41/06","H04L41/0654","G","G05","G05B","G05B9/00","G05B9/02","G05B9/03","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q50/00","G06Q50/10","H","H04","H04L","H04L63/00","H04L63/12","H04L63/123","H","H04","H04L","H04L67/00","H04L67/01","H04L67/12","H04L67/125","H","H04","H04L","H04L69/00","H04L69/08"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260060269A/en"},{"publication_number":"KR20260060270A","title":"Magnetic Tunnel Junction (MTJ) Device for Storing Ternary Weights and Neuromorphic Computing Array Utilizing the Same","abstract":"본 발명은 자유층(130), 터널 장벽층(120), 및 고정층(110)을 포함하는 자기터널접합(MTJ) 소자(100)에 있어서, 스핀 전달 토크(STT)에 의해 자유층(130)의 자화 방향이 고정층(110)의 자화 방향과 평행인 제1 상태(S1, 가중치 +1), 반평행인 제2 상태(S2, 가중치 -1), 및 직교하는 제3 상태(S3, 가중치 0)의 세 가지 안정 자화 상태를 가지며, 이 세 가지 상태가 각각 인공신경망의 삼진 가중치 {+1, -1, 0}에 대응하여 1.58비트 가중치를 저장하는 삼진 자기터널접합 소자(100) 및 이를 크로스바 어레이(200) 형태로 집적하여 입력 전압과 저항값의 곱셈 및 전류 합산에 의한 곱셈-누산 연산을 물리 법칙으로 수행하는 뉴로모픽 컴퓨팅 어레이(200)를 제공한다. The present invention provides a ternary magnetic tunnel junction (MTJ) device (100) comprising a free layer (130), a tunnel barrier layer (120), and a fixed layer (110), wherein the magnetization direction of the free layer (130) is parallel to the magnetization direction of the fixed layer (110) by spin transfer torque (STT), and there are three stable magnetization states: a first state (S1, weight +1) in which the magnetization direction of the free layer (130) is parallel to the magnetization direction of the fixed layer (110), a second state (S2, weight -1) in which it is antiparallel, and a third state (S3, weight 0) in which it is orthogonal. Each of these three states stores a 1.58-bit weight corresponding to the ternary weights {+1, -1, 0} of an artificial neural network. The invention also provides a neuromorphic computing array (200) that integrates the device in the form of a crossbar array (200) to perform multiplication-accumulation operations by multiplying input voltage and resistance values and summing currents using physical laws.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-05-04","filing_date":"2026-04-14","priority_date":"","cpc_codes":["H","H10","H10N","H10N50/00","H10N50/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G06N3/065"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260060270A/en"},{"publication_number":"KR102959116B1","title":"Paragraph reconstruction method and device for ai search","abstract":"본 발명은 AI 검색을 위한 문단 재구성 방법에 관한 것이다. AI 검색을 위한 문단 재구성 방법은, 문서를 분할하여 제1 문단을 생성하는 단계, 생성된 제1 문단을 제1 인공지능 모델에 제공하여 문맥 손실이 있는 적어도 하나의 영역을 추출하는 단계 및 문서의 내용을 기초로 추출된 적어도 하나의 영역과 연관된 문맥 보강을 수행하여 제1 문단이 재구성된 제2 문단을 생성하는 단계를 포함한다. The present invention relates to a method for reconstructing a paragraph for AI search. The method for reconstructing a paragraph for AI search comprises the steps of: dividing a document to generate a first paragraph; providing the generated first paragraph to a first artificial intelligence model to extract at least one region with context loss; and performing context reinforcement associated with the extracted at least one region based on the content of the document to generate a second paragraph in which the first paragraph is reconstructed.","assignee":"주식회사 로이드케이","inventors":["김대훈","김창수"],"publication_date":"2026-05-04","filing_date":"2025-12-31","priority_date":"","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/3332","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3347","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N20/00"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102959116B1/en"},{"publication_number":"KR102960361B1","title":"bag sewing defect detection system","abstract":"포대 재봉불량 검출시스템이 개시된다. 본 발명의 실시예에 따른 포대 재봉불량 검출시스템은, 포대 입구를 밀봉 처리하는 재봉기의 출구와 인접하여 장착되고, 밀봉 처리된 포대가 재봉기의 출구로부터 배출되는 것을 감지하는 포대 감지부; 재봉기의 출구와 인접하여 장착되고, 포대 감지부에 의해 포대가 감지될 경우 재봉기의 출구로부터 배출되는 포대 입구의 재봉상태를 촬영하는 영상 취득부; 상기 영상 취득부로부터 촬영된 촬영데이터를 바탕으로 재봉상태의 불량 여부를 판단하고, 불량 검출 알고리즘을 적용하여 불량 여부를 판단하며, 정상 형상 학습 및 비교 기준 생성 AI모델을 통해 불량 검출 알고리즘을 갱신함과 동시에 영상 취득부의 작동을 제어하는 영상분석 판정부; 및 상기 영상분석 판정부에 의해 불량 판정된 포대 검출 시 기설정된 경고신호를 출력하고, 불량 판정된 포대를 기설정된 방향으로 이송하도록 제어하는 불량포대 조치부;를 포함하는 것을 구성의 요지로 한다. A bag sewing defect detection system is disclosed. The bag sewing defect detection system according to an embodiment of the present invention comprises: a bag detection unit mounted adjacent to the exit of a sewing machine that seals the bag opening and detects that a sealed bag is discharged from the exit of the sewing machine; an image acquisition unit mounted adjacent to the exit of the sewing machine and photographs the sewing state of the bag opening discharged from the exit of the sewing machine when a bag is detected by the bag detection unit; an image analysis judgment unit that determines whether the sewing state is defective based on the photographic data captured by the image acquisition unit, determines whether the defect is defective by applying a defect detection algorithm, updates the defect detection algorithm through a normal shape learning and comparison standard generation AI model, and simultaneously controls the operation of the image acquisition unit; and a defective bag action unit that outputs a preset warning signal when a bag determined to be defective by the image analysis judgment unit is detected and controls the delivery of the defective bag in a preset direction.","assignee":"주식회사 대광이엘씨","inventors":["이성일"],"publication_date":"2026-05-04","filing_date":"2025-12-12","priority_date":"","cpc_codes":["D","D06","D06H","D06H3/00","D06H3/08","D","D05","D05B","D05B13/00","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/89","G01N21/892","G01N21/898","G01N21/8983","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102960361B1/en"},{"publication_number":"KR102960921B1","title":"Apparatus and method for managing population distribution based on habitat use characteristics of the population","abstract":"본 개시는 지역 별 패치 지표를 활용하여 서식지 이용 특성을 예측하고 개체군의 분포를 유도하는 시스템에 관련된 것이다. 본 개시에 따르면 개체군 분포 분석 장치는 대상 지역의 특징에 관한 서식 정보, 및 상기 대상 지역에 영향을 미치는 요인에 관한 교란 정보를 수집하고, 상기 서식 정보로부터, 단위 지역 별 특징을 지시하는 패치 지표 정보를 생성하고, AI 기반 서식 분석 모델을 활용하여, 상기 패치 지표 정보로부터 상기 대상 지역 내 포유류의 분포에 관련된 서식지 이용 특성을 예측하고, 상기 서식 정보, 상기 서식지 이용 특성, 및 상기 교란 정보를 이용하여, 목표 개체군 분포 조건을 결정하고, 상기 패치 지표 정보와 상기 목표 개체군 분포 조건에 기반하여 서식지 구조의 조정을 제안하는 서식지 조정 정보를 생성할 수 있다. The present disclosure relates to a system for predicting habitat use characteristics and inducing population distribution by utilizing regional patch indicators. According to the present disclosure, a population distribution analysis device collects habitat information regarding the characteristics of a target area and disturbance information regarding factors affecting the target area, generates patch indicator information indicating characteristics for each unit area from the habitat information, predicts habitat use characteristics related to the distribution of mammals within the target area from the patch indicator information by utilizing an AI-based habitat analysis model, determines target population distribution conditions using the habitat information, habitat use characteristics, and disturbance information, and generates habitat adjustment information that proposes adjustment of the habitat structure based on the patch indicator information and the target population distribution conditions.","assignee":"국립생태원","inventors":["황현수","신현철","권효정"],"publication_date":"2026-05-04","filing_date":"2025-11-21","priority_date":"","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/02","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06Q","G06Q50/00","G06Q50/10"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102960921B1/en"},{"publication_number":"LU603703B1","title":"Method and apparatus for predicting density of limestone slurry, storage medium, and electronic device","abstract":"Provided are a method and an apparatus for predicting density of limestone slurry, a storage medium, and an electronic device, relating to the technical field of wet desulfurization. The method includes: first, acquiring feature data related to the limestone slurry, where the feature data includes a liquid level of a limestone slurry tank, a current value of a limestone slurry pump, an instantaneous flow rate of the limestone slurry, and an instantaneous flow rate of limestone powder; and then, based on the feature data, predicting density of the limestone slurry by using a particle swarm optimization-support vector machine model. By applying the technical solution, based on the feature data related to the limestone slurry, the density of the limestone slurry is predicted in real time by using a particle swarm optimization-support vector machine algorithm, thereby improving the accuracy of measuring the density of the limestone slurry. (FIG. 1)","assignee":"Inner Mongolia Mengda Power Generation Co Ltd","inventors":["Haidong Yu"],"publication_date":"2026-05-04","filing_date":"2025-11-04","priority_date":"2024-11-18","cpc_codes":["G","G16","G16C","G16C20/00","G16C20/70","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G16","G16C","G16C20/00","G16C20/30"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU603703B1/en"},{"publication_number":"KR102961111B1","title":"Method of training deep learning models and device thereof","abstract":"본 개시의 일 실시예에 따르면, 사용자로부터 데이터셋을 입력받는 단계, 상기 데이터셋으로부터 하나 이상의 배치를 구성하는 단계, 상기 딥러닝 모델의 제1 초기 가중치 및 상기 하나 이상의 배치를 기반으로, 상기 딥러닝 모델을 학습시키는 단계, 상기 학습의 1회 에포크가 종료된 것에 대응하여, 가중치 업데이트 벡터를 산출하는 단계, 상기 데이터셋으로부터 클래스 균형 셋을 구성하는 단계, 상기 클래스 균형 셋의 상기 딥러닝 모델에 대한 손실 함수를 기반으로, 최적 리스케일링 계수를 탐색하는 단계, 및 상기 최적 리스케일링 계수, 상기 제1 초기 가중치, 및 상기 가중치 업데이트 벡터를 기반으로, 다음 에포크에 대한 제2 초기 가중치를 산출하는 단계를 포함하는 방법을 제공할 수 있다. According to one embodiment of the present disclosure, a method may be provided comprising the steps of: receiving a dataset from a user; forming one or more batches from the dataset; training the deep learning model based on a first initial weight of the deep learning model and the one or more batches; calculating a weight update vector in response to the completion of one epoch of the training; forming a class balance set from the dataset; searching for an optimal rescaling coefficient based on a loss function for the deep learning model of the class balance set; and calculating a second initial weight for the next epoch based on the optimal rescaling coefficient, the first initial weight, and the weight update vector.","assignee":"충남대학교 산학협력단","inventors":["장경선","이광희"],"publication_date":"2026-05-04","filing_date":"2025-11-03","priority_date":"","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102961111B1/en"},{"publication_number":"LU603642B1","title":"Prediction method for friction and wear properties of DLC coatings based on machine learning","abstract":"This invention discloses a method for predicting the friction and wear performance of DLC coatings based on machine learning, involving the interdisciplinary fields of surface engineering and machine learning. It includes collecting feature data and labeldata of DLC coatings, where the feature data includes image data, spectral data, and numerical data. The acquired data is cleaned and integrated to obtain a multimodal dataset. Based on the amount of data, a prediction model for the friction and wear performance of DLC coatings is constructed. Using the obtained multimodal dataset, the prediction model for the friction and wear performance of DLC coatings is trained, and the prediction performance of the model is evaluated using label data. Samples with large errors are analyzed to continuously optimize the prediction performance.","assignee":"Shenzhen Polytechnic Univ","inventors":["Zhijun Deng","Yanyan Lin","Zhurong Dong","Zhe Nie"],"publication_date":"2026-05-04","filing_date":"2025-10-30","priority_date":"2025-04-07","cpc_codes":["G","G01","G01N","G01N33/00","G01N33/0096","G","G01","G01N","G01N19/00","G01N19/02","G","G01","G01N","G01N21/00","G01N21/62","G01N21/63","G01N21/65","G","G01","G01N","G01N3/00","G01N3/40","G","G01","G01N","G01N3/00","G01N3/56","G","G06","G06N","G06N20/00","G","G01","G01N","G01N2203/00","G01N2203/0058","G01N2203/0069","G01N2203/0075"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU603642B1/en"},{"publication_number":"KR102960611B1","title":"Intelligent scent diffuser based on user emotions","abstract":"본 발명은 인공지능(AI) 기술과 디지털 센팅(Digital Scenting) 기술을 융합한 사용자 맞춤형 향기 제공 시스템에 관한 것으로, 보다 상세하게는 사용자의 실시간 감정 상태, 생체 신호 및 외부에서 제공되는 시청각 콘텐츠 데이터와 유기적으로 연동하여, 개인에게 최적화된 향기를 자동으로 추천, 조합 및 분사하고, 사용자의 피드백을 통해 스스로 학습하여 추천의 정확도를 지속적으로 향상시키는 사용자 감정 기반 지능형 향기 디퓨저 및 그 동작 방법에 관한 것이다. The present invention relates to a user-customized fragrance provision system that fuses artificial intelligence (AI) technology with digital scenting technology. More specifically, it relates to a user emotion-based intelligent fragrance diffuser and a method of operation thereof that organically links with the user's real-time emotional state, biosignals, and externally provided audiovisual content data to automatically recommend, combine, and dispense fragrances optimized for the individual, and continuously improve the accuracy of recommendations by self-learning through user feedback.","assignee":"(주)에센시아랩","inventors":["박주연","박형곤","이정선","홍광희","김기남","정경훈","정준원","이승민","김나리","김아람"],"publication_date":"2026-05-04","filing_date":"2025-10-29","priority_date":"2025-09-30","cpc_codes":["A","A61","A61L","A61L9/00","A61L9/14","G","G06","G06F","G06F3/00","G06F3/06","G06F3/08","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0631","H","H04","H04L","H04L67/00","H04L67/50","A","A61","A61L","A61L2209/00","A61L2209/10","A61L2209/11","A61L2209/111"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102960611B1/en"},{"publication_number":"KR102959512B1","title":"Smart food material demand forecasting method based on big data analysis results through deep learning-based artificial intelligence model and time-series data of food materials acquired from users","abstract":"스마트 식자재 수요 예측 방법이 개시된다. 스마트 식자재 수요 예측 방법은 통신 인터페이스를 이용하여, 사용자 단말로부터 대상 식자재에 대한 소정의 과거 시구간 동안의 시계열 데이터를 획득하고, 시계열 데이터에 대해 정규화 처리를 수행하여 가공 시계열 데이터를 생성한 후, 상기 가공 시계열 데이터를 메모리에 저장된 식자재 데이터베이스에 누적 저장하고, 가공 시계열 데이터를 딥러닝 기반 인공지능 모델에 입력하여, 현재 시점으로부터 소정의 미래 시구간 동안의 상기 대상 식자재에 대한 예측 수요량 추이를 획득하고, 상기 예측 수요량 추이, 상기 대상 식자재의 실시간 재고량 및 상기 대상 식자재의 상기 소정의 과거 시구간 동안의 평균 리드 타임(lead time)을 기반으로 상기 대상 식자재의 부족 위험도를 산출하고, 대상 식자재 중에서 상기 부족 위험도가 미리 설정된 임계 위험도 이상인 위험 식자재가 확인되면, 상기 위험 식자재에 대한 발주 필요 알림을 상기 사용자 단말에 제공하는 동작을 포함한다. A smart food ingredient demand forecasting method is disclosed. The smart food ingredient demand forecasting method includes the operation of using a communication interface to obtain time series data for a target food ingredient during a predetermined past time interval from a user terminal, performing normalization processing on the time series data to generate processed time series data, accumulating and storing the processed time series data in a food ingredient database stored in memory, inputting the processed time series data into a deep learning-based artificial intelligence model to obtain a predicted demand trend for the target food ingredient during a predetermined future time interval from the present time, calculating a shortage risk of the target food ingredient based on the predicted demand trend, the real-time inventory amount of the target food ingredient, and the average lead time of the target food ingredient during the predetermined past time interval, and, if a risk food ingredient is identified among the target food ingredients in which the shortage risk is greater than or equal to a preset threshold risk level, providing a notification to the user terminal that an order for the risk food ingredient is required.","assignee":"햇품 주식회사","inventors":["이남훈"],"publication_date":"2026-05-04","filing_date":"2025-10-24","priority_date":"","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087","G06Q10/0872","G06Q10/08726","G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087","G06Q10/0874","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087","G06Q10/0877","G","G06","G06F","G06F2123/00","G06F2123/02"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102959512B1/en"},{"publication_number":"KR20260059551A","title":"Method for providing image-based transferring service and banking server performing the same","abstract":"본 발명은 이미지 기반의 계좌이체 서비스 제공 방법 및 이를 수행하는 금융 서버를 개시한다. 상기 이미지 기반의 계좌이체 서비스 제공 방법은, 상기 사용자 단말로부터 계좌정보를 포함하는 이미지를 수신하는 단계, 상기 수신된 이미지를 인식 모듈에 입력하고, 상기 인식 모듈의 출력으로 상기 이미지에 포함된 문자에 대한 인식정보를 획득하는 단계, 금융거래 모듈로부터 이미 실행된 복수의 계좌이체내역에 해당하는 복수의 이체정보를 제공받고, 상기 이미지의 수신시간과 상기 복수의 이체정보 각각의 생성시간을 비교하여 상기 이미지에 대응되는 금융거래에 대한 이체정보를 추출하는 단계, 상기 인식정보 중 상기 이체정보에 대응되는 공통항목을 추출하고, 상기 인식정보와 상기 이체정보의 공통항목을 대상으로 로스값을 산출하는 단계 및 상기 로스값이 작아지도록 상기 인식 모듈을 학습시키는 단계를 포함할 수 있다. The present invention discloses a method for providing an image-based account transfer service and a financial server for performing the same. The method for providing an image-based account transfer service may include the steps of: receiving an image containing account information from a user terminal; inputting the received image into a recognition module and obtaining recognition information for characters included in the image as the output of the recognition module; receiving a plurality of transfer information corresponding to a plurality of account transfer details that have already been executed from a financial transaction module, and extracting transfer information for a financial transaction corresponding to the image by comparing the reception time of the image with the generation time of each of the plurality of transfer information; extracting a common item corresponding to the transfer information from the recognition information and calculating a loss value based on the common item between the recognition information and the transfer information; and training the recognition module so that the loss value is reduced.","assignee":"주식회사 카카오뱅크","inventors":["정성훈","곽영준"],"publication_date":"2026-04-30","filing_date":"2026-04-20","priority_date":"2023-09-12","cpc_codes":["G","G06","G06Q","G06Q20/00","G06Q20/08","G06Q20/10","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q20/00","G06Q20/38","G06Q20/40","G06Q20/401","G06Q20/4014","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V30/00","G06V30/40","G06V30/41","G","G08","G08B","G08B21/00","G08B21/18"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260059551A/en"},{"publication_number":"AU2026202728A1","title":"Wind-powered computing buoy","abstract":"Disclosed is a novel type of computing apparatus which is integrated within a buoy that obtains the energy required to power its computing operations from winds that travel across the surface of the body of water on which the buoy floats. Additionally, these self-powered computing buoys utilize their close proximity to a body of water in order to significantly lower the cost and complexity of cooling their computing circuits. Computing tasks of an arbitrary nature are supported, as is the incorporation and/or utilization of computing circuits specialized for the execution of specific types of computing tasks. And, each buoy’s receipt of a computational task, and its return of a computational result, may be accomplished through the transmission of data across satellite links, fiber optic cables, LAN cables, radio, modulated light, microwaves, and/or any other channel, link, connection, and/or network.","assignee":"Lone Gull Holdings Ltd","inventors":["Brian Lee Moffat","Garth Alexander SHELDON-COULSON"],"publication_date":"2026-04-30","filing_date":"2026-04-13","priority_date":"2018-01-27","cpc_codes":["B","B63","B63B","B63B22/00","B","B63","B63B","B63B22/00","B63B22/24","B","B63","B63B","B63B35/00","B63B35/44","B","B63","B63H","B63H13/00","B","B63","B63H","B63H21/00","B63H21/12","B63H21/17","F","F03","F03D","F03D13/00","F03D13/20","F03D13/25","F","F03","F03D","F03D9/00","F03D9/10","F","F03","F03D","F03D9/00","F03D9/10","F03D9/11","F","F03","F03D","F03D9/00","F03D9/30","F03D9/34","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","H","H04","H04L","H04L9/00","H04L9/06","H04L9/0643","H","H04","H04L","H04L9/00","H04L9/32","H04L9/3236","H04L9/3239","B","B63","B63B","B63B22/00","B63B2022/006","B","B63","B63B","B63B35/00","B63B35/44","B63B2035/4433","B63B2035/446","H","H04","H04L","H04L9/00","H04L9/50","Y","Y02","Y02E","Y02E10/00","Y02E10/30","Y","Y02","Y02E","Y02E10/00","Y02E10/70","Y02E10/727","Y","Y02","Y02E","Y02E70/00","Y02E70/30"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202728A1/en"},{"publication_number":"KR20260059539A","title":"AI processing system based on the integration of modular expert modules and method for managing the distribution of MoE expert modules","abstract":"본 발명은 사전 학습된 베이스 모델에 외부에서 개발된 도메인 특화 전문가 모듈의 가중치를 동적으로 로딩하여 결합하는 모듈형 전문가 결합 기반 인공지능 처리 시스템 및 MoE 전문가 모듈 유통 관리 방법을 개시한다. 본 발명의 시스템은 이종 전문가 모듈 간의 텐서 차원 불일치를 자동으로 해소하는 차원 투영 레이어를 포함하는 표준화된 입출력 인터페이스, 디지털 워터마크 기반 라이선스 검증을 수행하는 액세스 제어 모듈을 포함하는 모듈 결합부, 입력 데이터 속성에 따라 각 전문가 모듈의 출력 비중을 실시간으로 결정하는 적응형 게이팅 네트워크, 및 출력 분포 차이를 보정하는 연합 정규화 레이어를 포함한다. 본 발명에 따르면 AI 전문가 모듈의 가중치를 독립적 지식 자산으로서 유통ㆍ거래할 수 있는 기술적 기반이 제공되며, 기여도 기반 정산 시스템을 통해 공급자에게 기여 비율에 비례한 로열티가 자동 배분된다. The present invention discloses a modular expert combination-based artificial intelligence processing system and a method for managing the distribution of MoE expert modules, which dynamically load and combine weights of externally developed domain-specific expert modules into a pre-trained base model. The system of the present invention includes a standardized input/output interface comprising a dimension projection layer that automatically resolves tensor dimension mismatches between heterogeneous expert modules, a module combination unit comprising an access control module that performs digital watermark-based license verification, an adaptive gating network that determines the output weight of each expert module in real time according to input data attributes, and a federated normalization layer that corrects differences in output distribution. According to the present invention, a technical basis is provided for distributing and trading the weights of AI expert modules as independent knowledge assets, and royalties proportional to the contribution ratio are automatically distributed to suppliers through a contribution-based settlement system.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-30","filing_date":"2026-04-13","priority_date":"2026-04-13","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/10","G","G06","G06Q","G06Q50/00","G06Q50/10"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260059539A/en"},{"publication_number":"AU2026202709A1","title":"Automating the creation of listings using augmented reality computer technology","abstract":"Provided herein are system, apparatus, article of manufacture, method and/or computer program product embodiments, and/or combinations and sub-combinations thereof, for using computer technology to automate the creation of consistent, high quality listings for selling “for sale objects” (FSO) on an ecommerce site. Some embodiments are configured to or operate by: identifying a category of a FSO; accessing templates for the category from a template database; navigating a user through a computer generated augmented reality (AR) environment using the templates to generate images of the FSO; analyzing one or more of the images to determine characteristics of the FSO; and generating a listing for the FSO using the images and the characteristics. In some embodiments, the templates were generated from certain past listings of the category from a historical database, and wherein the certain past listings were selected based on one or more of: price achieved, time to sell, buyer feedback, sellability score, and/or difference between initial offer price and final selling price.","assignee":"Mercari Inc","inventors":["Byong Mok Oh"],"publication_date":"2026-04-30","filing_date":"2026-04-10","priority_date":"2019-09-16","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0641","G06Q30/0643","G","G06","G06F","G06F16/00","G06F16/90","G06F16/906","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0278","G","G06","G06T","G06T11/00","G","G06","G06T","G06T19/00","G06T19/003","G","G06","G06T","G06T19/00","G06T19/006","G","G06","G06T","G06T7/00","G06T7/0002","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V20/00","G06V20/20","G","G06","G06V","G06V20/00","G06V20/60","G06V20/62","G06V20/63"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202709A1/en"},{"publication_number":"AU2026202630A1","title":"Attention-based sequence transduction neural networks","abstract":"Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output sequence from an input sequence. In one aspect, one of the systems includes an encoder neural network configured to receive the input sequence and generate encoded representations of the network inputs, the encoder neural network comprising a sequence of one or more encoder subnetworks, each encoder subnetwork configured to receive a respective encoder subnetwork input for each of the input positions and to generate a respective subnetwork output for each of the input positions, and each encoder subnetwork comprising: an encoder self-attention sub-layer that is configured to receive the subnetwork input for each of the input positions and, for each particular input position in the input order: apply an attention mechanism over the encoder subnetwork inputs using one or more queries derived from the encoder subnetwork input at the particular input position. FIG. 1FIG. 1 WO 2018/217948 PCT/US2018/034224 1/3 Neural Network System 100 190 Output Attention-Based Neural Network 108 180 176 150 Add$Nom110 Format 174 Add $ Nom & Output Feed 134 Attention Sequence Forward Nx 152 130 172 & Nx Add % Attention Allention 170 132 Positional Positional Encoding Encoding Input Output 160 120 Inputs Outputs (shited right) Input Sequence 102 FIG. 1 20 26 20 26 30 08 A pr 2 02 6 W O 2 0 1 8 / 2 1 7 9 4 8 P C T / U S 2 0 1 8 / 0 3 4 2 2 4 2 0 2 6 2 0 2 6 2 0 2 6 3 0 0 8 A p r 1 5 0 A d d $ 1 1 0 1 7 4 $ N o m O u t p u t F e e d 1 3 4 N x 1 7 2 $ N o A d d % 1 7 0 1 3 2 O u t p u t 1 2 0 O u t p u t s 1 0 2","assignee":"Google LLC","inventors":["Aidan Nicholas GOMEZ","Llion Owen JONES","Lukasz Mieczyslaw Kaiser","Niki J. PARMAR","Illia POLOSUKHIN","Noam M. Shazeer","Jakob D. Uszkoreit","Ashish Teku VASWANI"],"publication_date":"2026-04-30","filing_date":"2026-04-08","priority_date":"2017-05-23","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202630A1/en"},{"publication_number":"KR20260057384A","title":"Systems and methods for identifying top alternative products based on a deterministic or inferential approach","abstract":"개시된 실시예는 사용자 쿼리에 기초하여 타깃 제품을 식별하고 대안 제품 추천을 생성하는 시스템 및 방법을 제공한다. 컴퓨터 구현 시스템은 기계 학습을 사용하여 사용자의 제품 모델 넘버 검색 쿼리와 연관된 복수의 속성 및 적어도 하나의 패턴을 결정하는 것을 포함하는 동작을 수행하도록 구성될 수 있다. 동작은 실험 데이터 세트에 기초하여 적어도 하나의 제품 카테고리 및 적어도 하나의 사용자에 의해 쿼리된 관심 제품을 결정하는 것을 더 포함할 수 있다. 동작은 쿼리된 관심 제품에 기초하여 타깃 제품을 결정하는 것을 더 포함할 수 있다. 동작은 실험데이터에 기초하여 쿼리된 제품과 연관된 복수의 주요 특징을 결정하고, 적어도 하나의 최고의 대안 제품을 결정하는 것을 더 포함할 수 있다. 동작은 외부 디바이스에 표시하기 위한 타깃 제품 및 최고의 대안 제품을 사용자에게 전송하는 것을 더 포함할 수 있다. The disclosed embodiments provide a system and method for identifying a target product and generating alternative product recommendations based on a user query. A computer-implemented system may be configured to perform an operation including using machine learning to determine a plurality of attributes and at least one pattern associated with a user's product model number search query. The operation may further include determining at least one product category and at least one product of interest queried by a user based on an experimental data set. The operation may further include determining a target product based on the queried product of interest. The operation may further include determining a plurality of key features associated with the queried product based on experimental data and determining at least one best alternative product. The operation may further include transmitting the target product and the best alternative product to the user for display on an external device.","assignee":"쿠팡 주식회사","inventors":["니산트 아그라왈","체탄 라오","에샨 헐리키"],"publication_date":"2026-04-28","filing_date":"2026-04-16","priority_date":"2022-03-25","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0631","G06Q30/06313","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0631","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9032","G06F16/90324","G06F16/90328","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0251","G06Q30/0255","G06Q30/0256","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0623","G06Q30/0625","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0641"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260057384A/en"},{"publication_number":"KR20260058195A","title":"Scenario Interaction Data Link Architecture for Compute-Efficient and Deterministic AI Agent Control in Game Environments","abstract":"본 발명은 대규모 게임 AI에이전트의 결정론적 제어를 위한 시나리오 상호작용 데이터링크 아키텍처(SIDLA)에 관한 것으로서, (1) 선행기술인 WL 기반 DGAE 구조를 게임 상태 인코딩에 응용한 결정론적 상태 인코더 (2) 2비트 이하 초저비트 양자화 추론 엔진 및 (3) 헤더 기반 JSON 패킷 통신 모듈(SIDLA 프로토콜)을 결합한다. 상기 구성에 의해GPU VRAM 점유를 기존 FP16 대비 1/10 이하로 감소시키고, 환각 발생률을 구조적으로 0%로 차단하며, 단일 서버에서 수천 개 이상의 AI 에이전트 NPC 인스턴스를 실시간으로 동시 구동하는 것이 가능하다. The present invention relates to a scenario interaction data link architecture (SIDLA) for deterministic control of a large-scale game AI agent, wherein (1) Deterministic state encoder that applies the prior art WL-based DGAE structure to game state encoding (2) Ultra-low bit quantization inference engine of 2 bits or less and (3) Combine header-based JSON packet communication module (SIDLA protocol). With the above configuration, GPU VRAM occupancy is reduced to less than 1/10 compared to existing FP16, the hallucination rate is structurally blocked to 0%, and it is possible to run thousands of AI agent NPC instances simultaneously in real time on a single server.","assignee":"이진우","inventors":["이진우"],"publication_date":"2026-04-28","filing_date":"2026-04-11","priority_date":"2026-04-11","cpc_codes":["A","A63","A63F","A63F13/00","A63F13/45","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","H","H04","H04L","H04L49/00","H04L49/90","H04L49/9057","H","H04","H04L","H04L69/00","H04L69/26","A","A63","A63F","A63F2300/00","A63F2300/50","A63F2300/51"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260058195A/en"},{"publication_number":"KR20260057593A","title":"Integrated Control System and Method for Operating and Supplying Energy to a Mobile Object","abstract":"본 발명은 이동체 운용 및 에너지 공급을 위한 통합 제어 시스템 및 그 방법에 관한 것이다. 본 발명에 따른 시스템은 이동체의 위치, 에너지 상태 및 이상 여부를 포함하는 상태 정보를 수집하는 감지부, 상기 상태 정보를 외부 네트워크를 통해 전송하는 통신부, 복수의 이동체 및 스테이션으로부터 수신된 상태 정보를 기반으로 이동체의 운용 가능 여부를 판단하는 외부 관제 서버, 상기 외부 관제 서버로부터 수신되는 승인 신호에 따라 이동체의 운용 또는 에너지 공급을 제어하는 제어부 및 비정상 상태 발생 시 동작을 차단하는 안전 차단부를 포함한다. 이에 따라, 이동체의 운용을 외부 관제 서버의 승인 기반으로 제어함으로써, 이동체 간 충돌 방지, 에너지 공급 최적화 및 통합 운용 효율 향상이 가능하다. The present invention relates to an integrated control system and method for operating a mobile body and supplying energy. The system according to the present invention includes a sensing unit that collects status information including the location, energy status, and abnormality of a mobile body; a communication unit that transmits the status information through an external network; an external control server that determines whether the mobile body can be operated based on status information received from a plurality of mobile bodies and stations; a control unit that controls the operation or energy supply of the mobile body according to an approval signal received from the external control server; and a safety blocking unit that blocks operation when an abnormal state occurs. Accordingly, by controlling the operation of the mobile body based on approval from the external control server, it is possible to prevent collisions between mobile bodies, optimize energy supply, and improve integrated operation efficiency.","assignee":"윤성민","inventors":["윤성민"],"publication_date":"2026-04-28","filing_date":"2026-04-10","priority_date":"2026-04-10","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","B","B25","B25J","B25J13/00","B25J13/006","B","B25","B25J","B25J5/00","B","B25","B25J","B25J9/00","B25J9/16","B25J9/1679","B","B60","B60L","B60L53/00","B60L53/30","B60L53/305","B","B60","B60L","B60L53/00","B60L53/60","B","B64","B64U","B64U20/00","B64U20/80","B64U20/87","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q50/00","G06Q50/06"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260057593A/en"},{"publication_number":"KR20260057367A","title":"Electrostatically trapped, laterally optically driven excitonic qubit device","abstract":"본 발명은 정전기적 트랩을 이용한 측방향 광 구동형 엑시토닉 큐비트 소자에 관한 것으로, 측방향에서 조사된 제어광을 수신하여 엑시톤을 생성하는 광생성층; 생성된 엑시톤이 수직 방향으로 이동하여 수용되는 양자 정보층; 양자 정보층 내의 엑시톤을 나노 스케일 영역에 구속하여 양자 점 상태를 형성하는 정전기적 게이트부; 및 구속된 엑시톤의 양자 상태 변화에 따른 방출광을 검출하는 광검출층을 포함하며, 게이트 전압에 의해 포텐셜 우물 깊이를 조절하여 큐비트의 초기화, 유지 및 제어를 수행한다. 측방향 광 입사 구조로 광 잡음 기인 탈위상이 억제되고, TMD 이종접합의 층간 엑시톤을 통해 결맞음 시간이 연장되며, 3차원 수직 적층 시 층별 독립 어드레싱이 가능하여 스케일 업이 용이하다. The present invention relates to a lateral light-driven excitonic qubit device using an electrostatic trap, comprising: a photogenerating layer that receives control light irradiated from the side and generates an exciton; a quantum information layer in which the generated exciton moves in a vertical direction and is received; an electrostatic gate portion that confines the exciton within the quantum information layer to a nanoscale region to form a quantum dot state; and a photodetector layer that detects emitted light according to a change in the quantum state of the confined exciton, and performs initialization, maintenance, and control of the qubit by adjusting the potential well depth by the gate voltage. With a lateral light incidence structure, dephase caused by optical noise is suppressed, the coherence time is extended through interlayer excitons of the TMD heterojunction, and independent addressing per layer is possible during 3D vertical stacking, making scaling up easy.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-28","filing_date":"2026-04-09","priority_date":"2026-04-09","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/40","B","B82","B82Y","B82Y10/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260057367A/en"},{"publication_number":"KR20260057375A","title":"AI Model Optimization System Across Heterogeneous Hardware via Static-Dynamic Calibration of Runtime Memory Structure Information","abstract":"본 발명은 GPU, TPU, NPU, IPU를 포함하는 이종 제1 하드웨어 아키텍처군에서 생성된 인공지능 모델을 ONNX를 매개로 특정 제2 하드웨어 아키텍처에서 실행하기 위한 모델 최적화 시스템에 관한 것이다. 본 발명의 하드웨어 분석 모듈은 제2 하드웨어 아키텍처의런타임 메모리 버스 토폴로지 및 캐시 라인 크기를 실시간으로 프로파일링하여 동적 실측값을 획득하고, 동적 실측값과 미리 도출된 정적 스펙값 간의 차이를 보정함으로써 실제 런타임 환경을 반영한 보정된 메모리 구조 정보를 생성한다. 최적화용 인공지능 엔진은 보정된 메모리 구조 정보를 입력으로 하여 각 레이어별 텐서 레이아웃 변환 및 메모리 타일링 크기를 자동으로 미세 조정함으로써 데이터 로드·스토어 병목 현상을 최소화한다. 본 발명에 따르면 발열, 클럭 저감, 버스 경합 등 런타임 환경에서 필연적으로 발생하는 정적-동적 메모리 특성 편차를 보정한 정보를 기반으로 최적화를 수행하므로, 정적 사양만을 활용하는 종래 최적화 도구 대비 현저히 향상된 실제 추론 성능이 달성된다. The present invention relates to a model optimization system for executing an artificial intelligence model generated in a heterogeneous first hardware architecture group including a GPU, TPU, NPU, and IPU on a specific second hardware architecture via ONNX. The hardware analysis module of the present invention profiles the runtime memory bus topology and cache line size of the second hardware architecture in real time to obtain dynamic actual values, and generates corrected memory structure information that reflects the actual runtime environment by correcting the difference between the dynamic actual values and pre-derived static specification values. The artificial intelligence engine for optimization minimizes data load and store bottlenecks by automatically fine-tuning the tensor layout transformation and memory tiling size for each layer using the corrected memory structure information as input. According to the present invention, since optimization is performed based on information that corrects static-dynamic memory characteristic deviations that inevitably occur in the runtime environment, such as heat generation, clock reduction, and bus contention, significantly improved actual inference performance is achieved compared to conventional optimization tools that utilize only static specifications.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-28","filing_date":"2026-04-09","priority_date":"2026-04-09","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260057375A/en"},{"publication_number":"KR20260056989A","title":"Exciton-based logic device with a lateral light incidence structure","abstract":"본 발명은 측방향 광 입사 구조를 갖는 엑시톤 기반 논리 소자에 관한 것이다. 본 발명의 소자는 소자의 측면으로 입사되는 제어광에 의해 엑시톤을 생성하는 광생성층, 광생성층의 하부에 배치되어 이동된 엑시톤이 방사 재결합하여 내부 방출광을 방출하는 중간층, 및 중간층의 하부에 배치되어 내부 방출광을 전기적 신호로 변환하는 광검출층을 포함한다. 제어광 조사 여부에 따라 광검출층의 출력 신호가 변화하여 비트 정보를 구현한다. 이차원 나노 물질을 이용한 Type-II 밴드 정렬 구조를 통해 상온에서도 엑시톤이 안정적으로 동작하며, 측방향 입사 구조는 소자의 3D 수직 적층 및 초고집적 배열을 가능하게 한다. 복수의 제어광 조합을 통해 AND, OR, XOR 등의 논리 게이트를 단일 소자 내에서 구현할 수 있다. The present invention relates to an exciton-based logic device having a lateral light incidence structure. The device of the present invention comprises a photogenerating layer that generates excitons by control light incident on the side of the device, an intermediate layer disposed below the photogenerating layer in which the moved excitons undergo radiative recombination to emit internally emitted light, and a photodetector layer disposed below the intermediate layer in which the internally emitted light is converted into an electrical signal. Bit information is implemented by changing the output signal of the photodetector layer depending on whether control light is irradiated. Through a Type-II band alignment structure using two-dimensional nanomaterials, excitons operate stably even at room temperature, and the lateral incidence structure enables 3D vertical stacking and ultra-high integration arrays of the device. Logic gates such as AND, OR, and XOR can be implemented within a single device through a combination of multiple control lights.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-27","filing_date":"2026-04-08","priority_date":"2026-04-08","cpc_codes":["H","H10","H10N","H10N70/00","H10N70/20","H10N70/257","G","G06","G06N","G06N10/00","G06N10/20","H","H10","H10N","H10N70/00","H10N70/801","H10N70/821","H10N70/826","H","H10","H10N","H10N70/00","H10N70/801","H10N70/881"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260056989A/en"},{"publication_number":"KR20260056986A","title":"Method, apparatus, and computer-readable storage medium for artificial intelligence inference using dynamic transition of floating-point formats based on routing scores in MoE architecture","abstract":"본 발명은 MoE(Mixture of Experts) 구조의 인공신경망에서, 라우팅 네트워크가 입력 데이터를 수신할 때마다 실시간으로 산출하는 전문가 모듈별 연산 중요도 점수에 기초하여, 각 전문가 모듈의 가중치 원본을 보존한 상태에서 부동소수점 포맷을 동적으로 선택하고 형변환하여 연산을 수행하는 인공지능 추론 방법, 장치 및 컴퓨터 판독 가능 저장매체에 관한 것이다. 연산 중요도 점수가 낮은 전문가 모듈에는 낮은 비트 수의 부동소수점 포맷을, 높은 전문가 모듈에는 높은 비트 수의 부동소수점 포맷을 할당함으로써 연산 자원을 차등 배분한다. 전문가 모듈의 캐시 적재 상태나 메모리 가용성과 무관하게 라우팅 출력 점수만을 기준으로 포맷을 결정하며, 동일한 전문가 모듈이라도 입력에 따라 상이한 부동소수점 포맷으로 연산이 수행된다. The present invention relates to an artificial intelligence inference method, apparatus, and computer-readable storage medium, wherein, in an artificial neural network with a Mixture of Experts (MoE) structure, a routing network performs computation by dynamically selecting and type-casting a floating-point format while preserving the original weights of each expert module, based on computational importance scores for each expert module calculated in real-time whenever input data is received. Computational resources are differentially allocated by assigning a low-bit floating-point format to expert modules with low computational importance scores and a high-bit floating-point format to expert modules with high computational importance scores. The format is determined based solely on the routing output score, regardless of the cache loading status or memory availability of the expert modules, and computation is performed using different floating-point formats depending on the input, even for the same expert module.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-27","filing_date":"2026-04-08","priority_date":"2026-04-08","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06F","G06F5/00","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/483","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260056986A/en"},{"publication_number":"KR20260056987A","title":"Layer-wise Importance-Score-based Variable Floating-Point Format Bit Allocation for AI Inference","abstract":"본 발명에 따른 인공지능 연산 방법은, 입력된 프롬프트 또는 데이터에 대하여 복수의 계층으로 구성된 인공신경망 내 각 계층의 연산 중요도 점수를 산출하는 단계와, 상기 산출된 연산 중요도 점수에 대응하여 각 계층의 가중치 연산에 적용될 부동소수점 포맷의 비트 수를 결정하는 단계와, 상기 결정된 부동소수점 포맷에 따라 각 계층의 연산을 수행하는 단계를 포함하며, 상기 비트 수 결정 단계는 연산 중요도 점수가 낮은 계층에는 낮은 비트 수의 부동소수점 포맷을, 연산 중요도 점수가 높은 계층에는 높은 비트 수의 부동소수점 포맷을 할당하는 것을 특징으로 한다. An artificial intelligence computation method according to the present invention comprises the steps of: calculating a computational importance score for each layer of an artificial neural network composed of multiple layers with respect to input prompts or data; determining the number of bits of a floating-point format to be applied to weighted computations of each layer in correspondence with the calculated computational importance score; and performing a computation of each layer according to the determined floating-point format, wherein the step of determining the number of bits is characterized by assigning a floating-point format with a low number of bits to a layer with a low computational importance score and a floating-point format with a high number of bits to a layer with a high computational importance score.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-27","filing_date":"2026-04-08","priority_date":"2026-04-08","cpc_codes":["G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/483","G","G06","G06F","G06F5/00","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/499","G06F7/49905","G06F7/4991","G06F7/49915","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260056987A/en"},{"publication_number":"KR20260056988A","title":"Layer-wise Importance-based Mixed-Precision Dynamic Quantization for AI Inference","abstract":"상기 목적을 달성하기 위한 본 발명에 따른 인공신경망의 연산 부하를 저감하기 위한 데이터 처리 방법은, 입력된 프롬프트 또는 데이터에 대하여 복수의 계층으로 구성된 인공신경망 내 각 계층의 연산 중요도를 산출하는 단계와, 상기 산출된 각 계층의 중요도에 대응하여 각 계층에 적용될 가중치의 비트 정밀도를 결정하는 단계와, 상기 결정된 비트 정밀도에 따라 부동소수점 기반의 가중치를 해당 비트 정밀도로 양자화하여 연산을 수행하는 단계를 포함하며, 상기 비트 정밀도 결정 단계는 중요도가 상대적으로 낮은 계층에는 1.58비트의 저정밀도를 할당하고, 중요도가 높아질수록 비트 정밀도 값을 단계적으로 증가시켜 할당하는 것을 특징으로 한다. A data processing method for reducing the computational load of an artificial neural network according to the present invention for achieving the above objective comprises: a step of calculating the computational importance of each layer within an artificial neural network composed of a plurality of layers for an input prompt or data; a step of determining the bit precision of a weight to be applied to each layer in correspondence with the calculated importance of each layer; and a step of performing a computation by quantizing a floating-point-based weight to the corresponding bit precision according to the determined bit precision. The bit precision determination step is characterized by allocating a low precision of 1.58 bits to layers with relatively low importance, and allocating a bit precision value that increases stepwise as importance increases.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-27","filing_date":"2026-04-08","priority_date":"2026-04-08","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/483"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260056988A/en"},{"publication_number":"KR20260056990A","title":"Exciton-based switching device","abstract":"본 발명은 엑시톤 기반 스위칭 소자에 관한 것으로, 측방향에서 조사된 광을 수신하여 엑시톤을 생성하는 광생성층(110); 상기 광생성층(110)의 하부에 배치되며, 이동된 엑시톤이 재결합되어 광을 방출하는 중간층(120); 및 상기 중간층(120)의 하부에 배치되며, 방출된 광을 검출하여 전기적 신호를 생성하는 광검출층(130)을 포함한다. 측방향 광 입사 구조에 의해 입력 광 경로와 출력 광 경로가 기하학적으로 분리되어 광 크로스토크가 억제되며, 엑시톤이라는 전기적 중성 준입자를 신호 매개체로 활용하여 전자기 간섭 없이 스위칭 동작이 구현된다. The present invention relates to an exciton-based switching device comprising: a photogenerating layer (110) that receives light irradiated from a side and generates an exciton; an intermediate layer (120) disposed below the photogenerating layer (110) in which the moved exciton recombines to emit light; and a photodetector layer (130) disposed below the intermediate layer (120) in which the emitted light is detected and an electrical signal is generated. By a lateral light incidence structure, the input light path and the output light path are geometrically separated to suppress optical crosstalk, and an electrically neutral quasiparticle called an exciton is utilized as a signal medium to enable switching operation without electromagnetic interference.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-27","filing_date":"2026-04-08","priority_date":"2026-04-08","cpc_codes":["H","H10","H10N","H10N70/00","H10N70/20","H10N70/257","G","G06","G06N","G06N10/00","G06N10/20","H","H10","H10F","H10F30/00","H10F30/20","H10F30/21","H10F30/28","H10F30/282","H","H10","H10N","H10N70/00","H10N70/801","H10N70/821","H10N70/826","H","H10","H10N","H10N70/00","H10N70/801","H10N70/881","H10N70/882"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260056990A/en"},{"publication_number":"KR20260056110A","title":"AI-based method for discovering scientific laws, system for performing the same, and computer-readable recording medium","abstract":"본 발명은 인공지능 시스템의 잠재 공간 내 엔트로피를 열역학적 상전이에 대응하는 4단계 방식으로 제어하여 미지의 과학 법칙을 자동으로 발견하는 방법에 관한 것이다. 본 발명은 관측 데이터와 기지의 물리 법칙을 이용하여 잠재 공간 내 논리적 기준점을 확정하는 저엔트로피 기준 확정 단계, 탐색 파라미터를 조정하여 고엔트로피 가설들을 광범위하게 생성하는 고엔트로피 가설 생성 단계, 시뮬레이션 환경과의 대조를 통해 가설을 정제하는 중간 엔트로피 정제 단계, 및 심볼릭 회귀와 파레토 최적화를 적용하여 최적 수식을 확정하는 신규 저엔트로피 수식 확정 단계를 포함한다. 각 단계 간 전환은 정량적 임계값에 의해 트리거되고, 선행 단계의 기준점이 후행 단계의 탐색 범위를 구속하는 계층적 제약 전파 구조를 통해 지식의 고착화를 방지하고 논리적 비약을 제어한다. The present invention relates to a method for automatically discovering unknown scientific laws by controlling the entropy within the potential space of an artificial intelligence system in a four-stage manner corresponding to thermodynamic phase transitions. The present invention includes a low-entropy reference determination step that determines a logical reference point within the potential space using observational data and known physical laws; a high-entropy hypothesis generation step that extensively generates high-entropy hypotheses by adjusting search parameters; an intermediate entropy refinement step that refines hypotheses through comparison with a simulation environment; and a novel low-entropy formula determination step that determines an optimal formula by applying symbolic regression and Pareto optimization. Transitions between each step are triggered by a quantitative threshold value, and knowledge stagnation and logical leaps are prevented through a hierarchical constraint propagation structure in which the reference point of a preceding step constrains the search range of a subsequent step.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-24","filing_date":"2026-04-07","priority_date":"2026-04-07","cpc_codes":["G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260056110A/en"},{"publication_number":"KR20260056111A","title":"Method and system for managing AI workflows using phase transition control in latent space","abstract":"본 발명은 거대언어모델(LLM)의 잠재 공간(Latent Space) 내 엔트로피(Entropy) 상태를 물질의 고체-액체-기체 상전이(엔트로피 천이)(Phase Transition) 원리에 대응시켜 단계적으로 제어함으로써, 수렴적 논리 확정에서 발산적 혁신 결과물까지를 체계적으로 도출하는 인공지능 워크플로우 관리 방법 및 시스템을 제공한다. 고체 엔트로피 단계(저엔트로피단계)에서는 낮은 온도(T) 파라미터로 LLM의 기준점(Ground Truth)과 논리적 골격을 확정하고, 액체 엔트로피 단계(중간 엔트로피 단계)에서는 온도와 탑-피(Top-p) 파라미터를 중간 수준으로 상향하여 다각적 대안을 생성하며, 기체 엔트로피 단계(고엔트로피단계)에서는 파라미터를 임계치 이상으로 극대화하여 고엔트로피 기반의 혁신적 결과물을 도출한다. 고체-액체-기체 각 단계에는 분석가, 설계자, 예술가 에이전트를 독립적으로 할당하는 다중 에이전트 구조가 적용되며, 기체 엔트로피 단계(고엔트로피단계) 결과물은 역상전이(엔트로피 천이)(Resolidification) 단계에서 제약 조건의 재적용을 통해 실현 가능한 최종 결과물로 정제된다. 또한, 실시간 엔트로피 측정부가 LLM의 출력 토큰 확률 분포를 분석하여 계산한 엔트로피 값이 임계치에 도달하면 다음 단계의 상전이(엔트로피 천이)를 자동으로 수행하는 자율 제어 구조를 포함한다. 본 발명은 LLM의 국소 최솟값(Local Minima) 함몰 문제를 해소하고, 현존하는 LLM API의 표준 파라미터만으로 즉시 구현 가능하며, 비즈니스 전략, 소프트웨어 개발, 연구개발, 특허 발명 지원 등 모든 지적 작업 영역에 범용적으로 적용될 수 있다. The present invention provides an artificial intelligence workflow management method and system that systematically derives results ranging from convergent logical determination to divergent innovation by controlling the entropy state within the latent space of a Large Language Model (LLM) in stages by correlating it with the principles of solid-liquid-gas phase transitions (entropy transitions) of matter. In the solid entropy stage (low entropy stage), the ground truth and logical framework of the LLM are determined using a low temperature (T) parameter; in the liquid entropy stage (intermediate entropy stage), multifaceted alternatives are generated by raising the temperature and Top-p parameters to an intermediate level; and in the gas entropy stage (high entropy stage), innovative results based on high entropy are derived by maximizing the parameters above a critical threshold. A multi-agent structure is applied in which analyst, designer, and artist agents are independently assigned to each of the solid, liquid, and gas phases. The output of the gas entropy phase (high-entropy phase) is refined into a feasible final result through the reapplication of constraints during the resolidification phase. Additionally, the system includes an autonomous control structure in which a real-time entropy measurement unit automatically performs the next phase transition when the entropy value calculated by analyzing the probability distribution of the LLM's output tokens reaches a threshold. This invention resolves the local minima collapse problem of the LLM, enables immediate implementation using only standard parameters of existing LLM APIs, and can be universally applied to all intellectual fields, including business strategy, software development, research and development, and patent invention support.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-24","filing_date":"2026-04-07","priority_date":"2026-04-07","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260056111A/en"},{"publication_number":"KR20260055383A","title":"Method and device for determining the priority of inspection","abstract":"검수 우선순위 결정 방법이 개시된다. 일 실시예에 따른 검수 우선순위 결정 방법은 입력 영상 및 입력 정보를 포함하는 입력 데이터를 획득하는 단계; 입력 영상을 인공 신경망 모델에 입력하여, 입력 영상에 대응하는 제1 스코어를 결정하는 단계; 통계 기반 알고리즘에 기초하여, 입력 정보에 대응하는 제2 스코어를 결정하는 단계; 제1 스코어 및 제2 스코어에 기초하여, 입력 데이터의 최종 스코어를 결정하는 단계; 및 최종 스코어에 기초하여, 입력 데이터의 검수 우선순위를 결정하는 단계를 포함할 수 있다. A method for determining inspection priority is disclosed. A method for determining inspection priority according to one embodiment may include: acquiring input data including an input image and input information; inputting the input image into an artificial neural network model to determine a first score corresponding to the input image; determining a second score corresponding to the input information based on a statistical-based algorithm; determining a final score of the input data based on the first score and the second score; and determining an inspection priority of the input data based on the final score.","assignee":"주식회사 카카오뱅크","inventors":["정기수","배정호","최호열","김수경","김판겸","강동훈"],"publication_date":"2026-04-23","filing_date":"2026-04-17","priority_date":"2022-12-01","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/95","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06F","G06F21/00","G06F21/30","G06F21/31","G06F21/33","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N7/00","G","G06","G06V","G06V10/00","G06V10/40","G06V10/42","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260055383A/en"},{"publication_number":"KR20260055377A","title":"Device and method for detecting abnormality","abstract":"본 발명은 사용자의 안면 이미지의 진위 여부를 판단할 수 있는 이상 탐지 장치에 관한 것이다. 상기 이상 탐지 장치는, 사용자의 안면 이미지를 수신하는 데이터 수집 모듈 및 수신된 상기 안면 이미지의 진위 여부에 관한 기대값을 산출하는 이미지 분석 모듈을 포함하되, 상기 이미지 분석 모듈은, 뉴럴 네트워크(Neural Network)를 포함하고, 수신된 상기 안면 이미지로부터 복수의 로짓(logit)을 추출하는 딥러닝부, 추출된 복수의 상기 로짓을 정규화하는 정규화부 및 정규화된 상기 로짓 및 상기 로짓에 할당된 이산화된 값을 이용하여 기대값을 산출하는 기대값 산출부를 포함한다. The present invention relates to an anomaly detection device capable of determining the authenticity of a user's facial image. The anomaly detection device comprises a data collection module that receives a user's facial image and an image analysis module that calculates an expected value regarding the authenticity of the received facial image, wherein the image analysis module includes a neural network and comprises a deep learning unit that extracts a plurality of logits from the received facial image, a normalization unit that normalizes the extracted plurality of logits, and an expected value calculation unit that calculates an expected value using the normalized logits and discretized values assigned to the logits.","assignee":"주식회사 카카오뱅크","inventors":["곽영준"],"publication_date":"2026-04-23","filing_date":"2026-04-15","priority_date":"2022-07-04","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/95","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06V","G06V10/00","G06V10/20","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G06V40/161","G","G06","G06V","G06V40/00","G06V40/40"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260055377A/en"},{"publication_number":"AU2026202572A1","title":"Systems and methods for managing agnostic data forms for vendors","abstract":"System and methods are provided for achieving data standardization and normalization through an Agnostic Data Format (ADF) architecture. ADFs systems and processes provide a transformative bridge, enabling disparate data sources to converge into a unified and standardized format within the Real-Time Data Mesh (RTDM) framework. This dynamic process utilizes Artificial Intelligence (AI) and Machine Learning (ML) algorithms to interpret and align diverse data attributes. The ADF management system, integrated into a dynamic event-driven architecture, allows vendors to interact with RTDM by translating and standardizing their data. The synchronized data integrates canonically, incorporating real- time updates and collaborative decision-making across the distribution platform. This innovative approach enhances operational efficiency, enables data-driven decision-making, and provides users improved ability to use data within the distribution ecosystem. 20 26 20 25 72 07 A pr 2 02 6 2 0 2 6 2 0 2 5 7 2 0 7 A p r 2 0 2 6","assignee":"Ingram Micro Inc","inventors":["Sanjib Sahoo"],"publication_date":"2026-04-23","filing_date":"2026-04-07","priority_date":"2024-02-21","cpc_codes":["H","H04","H04L","H04L67/00","H04L67/01","H04L67/12","G","G06","G06F","G06F16/00","G06F16/20","G06F16/25","G06F16/258","G","G06","G06F","G06F16/00","G06F16/20","G06F16/23","G06F16/2379","G","G06","G06F","G06F16/00","G06F16/20","G06F16/25","G06F16/254","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G","G06","G06Q","G06Q10/00","G06Q10/08","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087","G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/109","G","G06","G06Q","G06Q30/00","G06Q30/01","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0202","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0283"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202572A1/en"},{"publication_number":"KR20260055346A","title":"AI-based multi-stage professional translation system utilizing organic integration of NMT, LLM, and LRM","abstract":"본 발명은 NMT, LLM 및 LRM의 유기적 결합을 이용한 인공지능 기반 다단계 전문 번역 시스템에 관한 것이다. 본 발명의 시스템(100)은 원문 데이터를 입력받는 입력부(110), 기설정된 용어 사전(122) 및 번역 메모리(123)에 기초하여 NMT 방식으로 용어 일관성이 유지된 제1 번역물을 생성하는 제1 번역 모듈(120), 사용자의 프롬프트 지시어에 기초하여 LLM 방식으로 제1 번역물의 문맥 및 어조를 재구성하여 제2 번역물을 생성하는 제2 번역 모듈(130), 및 LRM 방식으로 제2 번역물과 원문 데이터 간의 기술적 논리 정합성을 분석하여 논리적 오류를 자동 교정하고 최종 번역물을 생성하는 제3 번역 모듈(140)을 포함한다. 제3 번역 모듈(140)은 원문으로부터 논리 구조 맵(Logic Structure Map)을 추출하고, 번역문과의 대조를 통해 수치 역전, 인과 방향 역전, 조건 누락, 구성요소 참조 단절 등의 논리 오류를 유형별로 자동 교정한다. 피드백 루프 모듈(150)은 LRM의 정합성 점수를 기준으로 고품질 번역 단위만을 선택적으로 번역 메모리(123)에 환류하는 자기강화 루프를 구현한다. 본 발명은 폐쇄형 서버(180) 환경에서 동작하여 데이터 보안을 보장하며, 특허, 의학, 법률 등 고도의 논리 무결성이 요구되는 전문 문서 번역에서 현저한 번역 품질 향상 효과를 제공한다. The present invention relates to an artificial intelligence-based multi-stage professional translation system utilizing an organic combination of NMT, LLM, and LRM. The system (100) of the present invention includes an input unit (110) that receives source text data, a first translation module (120) that generates a first translation with term consistency maintained in the NMT manner based on a pre-set term dictionary (122) and a translation memory (123), a second translation module (130) that generates a second translation by reconstructing the context and tone of the first translation in the LLM manner based on a user's prompt instruction, and a third translation module (140) that generates a final translation by automatically correcting logical errors by analyzing the technical logical consistency between the second translation and source text data in the LRM manner. The third translation module (140) extracts a Logic Structure Map from the source text and automatically corrects logical errors by type, such as numerical inversion, causal direction inversion, omission of conditions, and disconnection of component references, through comparison with the translation text. The feedback loop module (150) implements a self-reinforcing loop that selectively recirculates only high-quality translation units to the translation memory (123) based on the consistency score of the LRM. The present invention operates in a closed server (180) environment to ensure data security and provides a significant improvement in translation quality in the translation of specialized documents requiring high logical integrity, such as patents, medicine, and law.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-23","filing_date":"2026-04-03","priority_date":"2026-04-03","cpc_codes":["G","G06","G06F","G06F40/00","G06F40/40","G06F40/58","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260055346A/en"},{"publication_number":"AU2026202531A1","title":"Systems and methods for automating clinical workflow decisions and generating a priority read indicator","abstract":"Attorney Docket No. 04576.0100WOU1 / BSH.0124 Examples of the present disclosure describe systems and methods for automating clinical workflow decisions. In aspects, patient data may be collected from multiple data sources, such as patient records, imaging data, etc. The patient data may be processed using an artificial intelligence (AI) component. The output of the AI component may be used by healthcare professionals to inform healthcare decisions for patients. The output of the AI component and additional information relating to the healthcare decisions and healthcare paths may be provided as input to a decision analysis component. The decision analysis component may process the input and output an automated healthcare recommendation that may be used to further inform the healthcare decisions of the healthcare professionals. In some aspects, the output of the decision analysis component may be used to determine a priority or timeline for performing one or more actions relating to patient healthcare.","assignee":"Hologic Inc","inventors":["Biao Chen","Haili Chui","Nikolaos Gkanatsios","Zhenxue Jing","Ashwini Kshirsagar"],"publication_date":"2026-04-23","filing_date":"2026-04-02","priority_date":"2019-07-31","cpc_codes":["A","A61","A61B","A61B6/00","A61B6/02","A61B6/025","A","A61","A61B","A61B6/00","A61B6/50","A61B6/502","A","A61","A61B","A61B6/00","A61B6/52","A61B6/5211","A61B6/5217","A","A61","A61B","A61B8/00","A61B8/08","A61B8/0825","A","A61","A61B","A61B8/00","A61B8/52","A61B8/5215","A61B8/5223","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06316","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G06T7/0014","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H20/00","G16H20/40","G","G16","G16H","G16H30/00","G16H30/20","G","G16","G16H","G16H30/00","G16H30/40","G","G16","G16H","G16H40/00","G16H40/20","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/70","G","G16","G16H","G16H70/00","G16H70/20","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10132","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30004","G06T2207/30068","G","G16","G16H"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202531A1/en"},{"publication_number":"KR20260055345A","title":"Device, method and program for predicting mixture properties based on artificial intelligence","abstract":"복수의 물질로 구성된 혼합물의 물성을 예측하는 장치가 개시된다. 상기 장치는, 물질 정보에 대한 제1 특징 데이터를 출력하도록 학습된 제1 AI 모델과, 상기 제1 특징 데이터에 대한 물성 예측 정보를 출력하도록 학습된 제2 AI 모델이 저장된 메모리 및 상기 제1 AI 모델 및 상기 제2 AI 모델을 실행하는 프로세서를 포함하고, 상기 프로세서는, 상기 복수의 물질 각각에 대한 물질 정보를 상기 제1 AI 모델에 입력하여 상기 복수의 물질 각각에 대한 제1 특징 데이터를 획득하고; 상기 제1 특징 데이터를 상기 제2 AI 모델에 입력하여 상기 혼합물에 대한 물성 예측 정보를 획득할 수 있다. An apparatus for predicting the physical properties of a mixture composed of multiple substances is disclosed. The apparatus includes a memory in which a first AI model trained to output first feature data regarding substance information and a second AI model trained to output physical property prediction information regarding the first feature data are stored, and a processor that executes the first AI model and the second AI model. The processor can input substance information regarding each of the multiple substances into the first AI model to obtain first feature data regarding each of the multiple substances; and input the first feature data into the second AI model to obtain physical property prediction information regarding the mixture.","assignee":"주식회사 Lg 경영개발원; 주식회사 엘지에너지솔루션","inventors":["박창영","양홍준","이재완","한세희","전혜림","이보람","이창훈","김형태"],"publication_date":"2026-04-23","filing_date":"2026-04-01","priority_date":"2022-10-18","cpc_codes":["G","G16","G16C","G16C60/00","G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G16","G16C","G16C20/00","G16C20/20","G","G16","G16C","G16C20/00","G16C20/30","G","G16","G16C","G16C20/00","G16C20/70"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260055345A/en"},{"publication_number":"KR20260054671A","title":"An Efficient Parallel Computing Architecture using Phase Synchronization and Energy Balance Resonance and Method for Controlling the Same","abstract":"본 발명은 기존 양자 컴퓨터의 극저온 유지 한계를 극복하기 위해, 거시적 물리 현상인 '3상 평형 에너지' 모델과 '회전 위상 공명' 메커니즘을 결합한 양자 연산 모사 위상 공명 기반 아키텍처 및 그 제어 방법에 관한 것이다. 본 아키텍처는 입력 데이터를 평탄화한 후, 120도의 위상차를 갖는 3개 성분의 합산 에너지를 일정하게 유지하여 노이즈 저항성을 확보한다. 특히, 3상 중 하나를 연산에서 배제하여 남은 2개 위상 관계로부터 디지털 비트를 확정하는 '관측 기반 비트 생성' 로직을 통해 파동의 중첩 상태에서 유효 데이터를 추출한다. 또한, 엔진의 에너지가 피크(Peak)에 도달하는 특정 위상 마디(64괘 인덱스)에서만 샘플링을 수행하고 잔상을 폐기하는 **'필름 컷(Film-cut) 방식의 스파스 컴퓨팅'**을 적용함으로써, 고전적 전수 조사 대비 연산 효율을 지수적으로 높이고 전력 소모를 획기적으로 절감한다. 이는 파동의 간섭을 이용한 '확률적 수렴' 구조를 통해 비인가 접근 시 위상을 즉각 붕괴(Phase Collapse)시키는 능동 방어 기능을 제공하여 데이터 기밀성을 극대화한다. The present invention relates to a quantum computation simulation phase resonance-based architecture and a control method thereof, which combines a 'three-phase equilibrium energy' model, a macroscopic physical phenomenon, and a 'rotational phase resonance' mechanism to overcome the cryogenic maintenance limitations of existing quantum computers. This architecture ensures noise resistance by flattening the input data and maintaining a constant sum of energy of three components with a phase difference of 120 degrees. In particular, it extracts valid data from a superposition state of waves through an 'observation-based bit generation' logic that excludes one of the three phases from the calculation and determines digital bits from the remaining two phase relationships. In addition, by applying **'film-cut sparse computing'**, which performs sampling only at specific phase nodes (64-hexagram index) where the engine's energy reaches a peak and discards residual images, computational efficiency is increased exponentially compared to classical exhaustive scanning and power consumption is drastically reduced. This maximizes data confidentiality by providing an active defense function that immediately collapses the phase upon unauthorized access through a 'stabilistic convergence' structure utilizing wave interference.","assignee":"조익연","inventors":["조익연"],"publication_date":"2026-04-22","filing_date":"2026-04-05","priority_date":"2026-04-05","cpc_codes":["G","G06","G06N","G06N99/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260054671A/en"},{"publication_number":"KR20260053259A","title":"Electronic apparatus and method for controlling thereof","abstract":"전자 장치가 개시된다. 전자 장치는 통신 인터페이스, 메모리 및 프로세서를 포함한다. 본 개시에 따른 프로세서는, 날씨 정보 및 에어컨이 설치된 공간에 대한 정보를 획득하고, 날씨 정보 및 상기 공간에 대한 정보를 바탕으로 신경망 모델을 학습하고, 통신 인터페이스를 통해 에어컨의 운전 정보 및 상기 공간의 측정된 온도가 획득되면, 상기 공간의 측정된 온도 및 외부 온도를 신경망 모델에 입력하여 상기 공간에 대한 시간 별 예측 온도 정보를 획득하고, 예측 온도 정보 및 상기 공간의 측정된 온도를 바탕으로 에어컨에 결함이 존재하는지 여부를 판단하고, 에어컨에 결함이 존재하면, 알림 신호를 생성한다. An electronic device is disclosed. The electronic device includes a communication interface, a memory, and a processor. A processor according to the present disclosure acquires weather information and information about a space in which an air conditioner is installed, trains a neural network model based on the weather information and the information about the space, and when operation information of the air conditioner and the measured temperature of the space are acquired through the communication interface, inputs the measured temperature of the space and the external temperature into the neural network model to acquire hourly predicted temperature information for the space, determines whether a defect exists in the air conditioner based on the predicted temperature information and the measured temperature of the space, and if a defect exists in the air conditioner, generates a notification signal.","assignee":"삼성전자주식회사","inventors":["송성근","김경재","조혜정","이제헌","송관우"],"publication_date":"2026-04-21","filing_date":"2026-04-06","priority_date":"2020-08-20","cpc_codes":["F","F24","F24F","F24F11/00","F24F11/30","F24F11/32","F24F11/38","F","F24","F24F","F24F11/00","F24F11/50","F24F11/52","F","F24","F24F","F24F11/00","F24F11/50","F24F11/56","F","F24","F24F","F24F11/00","F24F11/62","F24F11/63","F","F24","F24F","F24F11/00","F24F11/62","F24F11/63","F24F11/64","G","G06","G06N","G06N3/00","G06N3/02","F","F24","F24F","F24F11/00","F24F11/50","F24F11/56","F24F11/58","F","F24","F24F","F24F2110/00","F24F2110/10","F","F24","F24F","F24F2110/00","F24F2110/10","F24F2110/12","F","F24","F24F","F24F2130/00","F24F2130/10","F","F24","F24F","F24F2140/00","F","F24","F24F","F24F2140/00","F24F2140/60"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260053259A/en"},{"publication_number":"KR20260054040A","title":"System and Method for Variable Control of Artificial Intelligence Visual Personas Linked to Real-Time Conversational Context Based on User-Defined Prompts","abstract":"본 발명은 사용자 정의 프롬프트 기반의 실시간 대화 맥락 연동형 인공지능 시각적 페르소나 가변 제어 시스템 및 방법에 관한 것이다. 본 발명에 따른 방법은 사용자로부터 기초 설정 프롬프트(400)를 수신하고, 이를 기반으로 기초 개체 이미지(500)를 생성하며, 라이브 통화 세션(800) 중 실시간 대화 데이터로부터 동적 컨텍스트(600)를 추출하고, 상기 기초 개체 이미지(500)의 시각적 일관성을 유지하면서 동적 컨텍스트(600)에 대응하는 세부 속성을 실시간으로 변형하여 업데이트된 개체 이미지(700)를 생성하고 출력하는 것을 특징으로 한다. 본 발명은 사용자가 정의한 인공지능의 핵심 시각적 정체성을 고정 파라미터로 보존하면서, 대화 맥락에 따른 가변 파라미터만을 실시간으로 조정하는 이중 파라미터 구조를 통해, 시각적 일관성과 동적 맥락 반응성을 동시에 달성하는 효과를 제공한다. The present invention relates to a system and method for controlling a variable artificial intelligence visual persona linked to real-time conversation context based on a user-defined prompt. The method according to the present invention is characterized by receiving a basic setting prompt (400) from a user, generating a basic object image (500) based thereon, extracting a dynamic context (600) from real-time conversation data during a live call session (800), and generating and outputting an updated object image (700) by modifying detailed attributes corresponding to the dynamic context (600) in real-time while maintaining the visual consistency of the basic object image (500). The present invention provides the effect of simultaneously achieving visual consistency and dynamic context responsiveness through a dual parameter structure that preserves the core visual identity of the user-defined artificial intelligence as a fixed parameter while adjusting only variable parameters according to the conversation context in real-time.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-21","filing_date":"2026-04-04","priority_date":"2026-04-04","cpc_codes":["G","G10","G10L","G10L15/00","G10L15/22","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G10","G10L","G10L15/00","G10L15/08","G10L15/18"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260054040A/en"},{"publication_number":"KR20260054041A","title":"System and Method for User-Selectable Multi-Mode Interface for Providing Visual Personas of Conversational Artificial Intelligence","abstract":"본 발명은 대화형 인공지능의 시각적 페르소나를 제공하기 위한 사용자 선택형 멀티 모드 인터페이스 시스템 및 방법에 관한 것으로서, 라이브통화 세션에서 인공지능 개체 이미지의 생성 방식을 사용자 정의 고정형(제1 모드), 맥락 기반 자동 생성형(제2 모드), 하이브리드 가변형(제3 모드)의 세 가지 독립적 모드로 구조화하고, 사용자 단말(100)의 모드 선택 인터페이스 표시부(140)를 통해 사용자가 원하는 모드를 능동적으로 선택하거나 세션 도중 실시간으로 전환할 수 있도록 하며, 모드별 최적화된 UI(610, 620, 630), 네트워크 상태 기반 동적 리소스 조절(520), 및 과거 이력 기반 개인화 모드 추천(260)을 통해 다양한 사용자 요구와 환경에 최적화된 AI 시각적 페르소나 경험을 제공한다. The present invention relates to a user-selectable multi-mode interface system and method for providing a visual persona of conversational artificial intelligence. The method of generating an artificial intelligence object image in a live call session is structured into three independent modes: a user-defined fixed type (first mode), a context-based automatic generation type (second mode), and a hybrid variable type (third mode). The user can actively select the desired mode or switch to it in real time during the session through a mode selection interface display unit (140) of a user terminal (100). Furthermore, the invention provides an AI visual persona experience optimized for various user needs and environments through a mode-optimized UI (610, 620, 630), network status-based dynamic resource control (520), and history-based personalized mode recommendation (260).","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-21","filing_date":"2026-04-04","priority_date":"2026-04-04","cpc_codes":["G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G06F16/9535","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06T","G06T13/00","G06T13/20","G06T13/40","H","H04","H04L","H04L51/00","H04L51/02"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260054041A/en"},{"publication_number":"KR20260054032A","title":"System and Method for Controlling Embedding Space Weights Between Heterogeneous Language Models to Protect Domain-Specific Data and Enhance Specialized Expertise","abstract":"본 발명은 기업 내부 데이터를 학습한 제1 모델(Private Model)(110)과 상용 거대 언어 모델인 제2 모델(Public Model)(120)의 임베딩 벡터(V1, V2)를 동적으로 융합하는 임베딩 융합 시스템(100)에 관한 것이다. 상기 시스템(100)은, 사용자 질의 내 도메인 키워드를 NER 및 IDF 기반으로 식별하는 키워드 분석 모듈(130), 키워드의 도메인 온톨로지(200) 매핑 스코어ㆍ키워드 밀도ㆍ의미 그래프 중심성을 복합적으로 분석하여 어텐션 가중치 팩터(á, â)를 실시간 산출하는 가중치 산출 모듈(140), 학습 가능한 선형 투영 행렬로 이종 차원 벡터를 정렬한 후 볼록 결합으로 통합 잠재 벡터(Vf)를 생성하는 공간 융합 함수 실행 모듈(150), 어텐션 가중치(á)가 임계값 초과 시 마스킹 및 노이즈 주입으로 역추론을 차단하는 데이터 보호 레이어(160), 신뢰도 스코어 기반 실시간 재추론을 수행하는 피드백 루프 모듈(170), 및 강화 학습으로 도메인 온톨로지(200)를 자율 갱신하는 매핑 자율 갱신부(190)를 포함한다. 본 발명에 의하면, 기업 전문 도메인 질의에 대한 답변 정확도 향상, 기업 기밀의 수학적 보호, 및 도메인 지식의 자율 적응 갱신이 달성된다. The present invention relates to an embedding fusion system (100) that dynamically fuses embedding vectors (V1, V2) of a first model (Private Model) (110) that has learned internal corporate data and a second model (Public Model) (120) that is a commercial large language model. The above system (100) includes a keyword analysis module (130) that identifies domain keywords within a user query based on NER and IDF, a weight calculation module (140) that calculates attention weight factors (á, â) in real time by comprehensively analyzing the domain ontology (200) mapping score, keyword density, and semantic graph centrality of the keywords, a spatial fusion function execution module (150) that aligns heterogeneous dimension vectors with a learnable linear projection matrix and then generates an integrated latent vector (Vf) through convex combination, a data protection layer (160) that blocks back-inference by masking and noise injection when the attention weight (á) exceeds a threshold, a feedback loop module (170) that performs real-time re-inference based on confidence scores, and a mapping autonomous update unit (190) that autonomously updates the domain ontology (200) through reinforcement learning. According to the present invention, improvement of answer accuracy for corporate specialized domain queries, mathematical protection of corporate secrets, and autonomous adaptive updating of domain knowledge are achieved.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-21","filing_date":"2026-04-01","priority_date":"2026-04-01","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/3332","G06F16/3334","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3346","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3347","G","G06","G06F","G06F16/00","G06F16/30","G06F16/36","G06F16/367","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260054032A/en"},{"publication_number":"KR20260052016A","title":"Method for fast distributed data processing to develop Artificial Intelligence model for image recognition of power facilities","abstract":"독립적인 환경에서 나타나는 자원관리의 문제점, 분석 결과 공유의 문제점과 전문 도메인지식의 필요로 하여 나타나는 개발방식의 단점을 해결하기 위한 AI(Artificial Intelligence) 모델 고속 분산 처리 시스템이 개시된다. 상기 AI 모델 고속 분산 처리 시스템은, 전력설비 이미지의 인식을 위한 AI(Artificial Intelligence) 모델을 산출하는 분석 환경을 제공하는 AI 분산 프레임워크, 산출되는 상기 AI 모델에 대해 미리 설정되는 분산환경에 따라 분산학습을 통해 상기 AI 모델의 유효성을 검증하는 분산학습 실행 환경부, 및 상기 전력설비 이미지를 갖는 전력설비 이미지 데이터, 및 상기 AI 모델을 저장하는 데이터 처리 프레임워크를 포함하는 것을 특징으로 한다. An AI (Artificial Intelligence) model high-speed distributed processing system is disclosed to resolve the problems of resource management arising in independent environments, the problems of sharing analysis results, and the disadvantages of development methods arising from the need for specialized domain knowledge. The AI model high-speed distributed processing system is characterized by comprising: an AI distributed framework that provides an analysis environment for generating an AI (Artificial Intelligence) model for recognizing power facility images; a distributed learning execution environment unit that verifies the validity of the AI model through distributed learning according to a distributed environment pre-configured for the generated AI model; power facility image data having the power facility images; and a data processing framework that stores the AI model.","assignee":"한국전력공사","inventors":["임채상","송찬호","허성오","이정일"],"publication_date":"2026-04-17","filing_date":"2026-04-08","priority_date":"2021-08-26","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/7715","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/776","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260052016A/en"},{"publication_number":"CO2026004663A2","title":"A system and method for identifying a variation of a phrase in a text passage","abstract":"RESUMEN Un sistema y método para identificar la ocurrencia de una variación semántica de una frase en un pasaje por al menos un procesador puede incluir calcular un vector de incrustación de frases, que representa un significado semántico de la frase; extraer, de una representación textual del pasaje, al menos un conjunto jerárquico de secuencias anidadas de palabras; para cada secuencia, calcular un vector de incrustación de secuencias correspondiente, que representa un significado semántico de la secuencia; para uno o más vectores de incrustación de secuencias, calcular un valor de similitud vectorial correspondiente, que representa la similitud de los vectores de incrustación de secuencias con el vector de incrustación de frases; identificar una secuencia correspondiente a un valor máximo de similitud vectorial del uno o más valores de similitud vectorial; y determinar la secuencia identificada como una variación semántica de la frase, basado en el valor máximo de similitud vectorial. ABSTRACT A system and method for identifying the occurrence of a semantic variation of a phrase in a passage by at least one processor may include calculating a phrase embedding vector, which represents a semantic meaning of the phrase; extracting, from a textual representation of the passage, at least one hierarchical set of nested word sequences; for each sequence, calculating a corresponding sequence embedding vector, which represents a semantic meaning of the sequence; for one or more sequence embedding vectors, calculating a corresponding vector similarity value, which represents the similarity of the sequence embedding vectors to the phrase embedding vector; identifying a sequence corresponding to a maximum vector similarity value from the one or more vector similarity values; and determining the sequence identified as a semantic variation of the phrase, based on the maximum vector similarity value.","assignee":"Genesys Cloud Services Inc","inventors":["Avraham Faizakof","Lev Haikin","Eyal Orbach","Nelly David","Rotem Moaz"],"publication_date":"2026-04-16","filing_date":"2026-04-10","priority_date":"2023-10-10","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/10","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/289","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N5/00","G06N5/04"],"country":"CO","kind":"application","source_url":"https://patents.google.com/patent/CO2026004663A2/en"},{"publication_number":"KR20260051318A","title":"Bit pattern operation method and operator using dynamic bit shift","abstract":"본 개시는 연산기에 의해 수행되는, 동적 비트 시프트를 이용한 비트패턴 연산 방법에 관한 것이다. 동적 비트 시프트를 이용한 비트패턴 연산 방법에 있어서, 제1 피연산자와 연관된 제1 비트패턴과 제2 피연산자와 연관된 제2 비트패턴의 지수 정렬(align)을 위한 제1 시프트(shift)의 수행 여부를 결정하는 단계, 제1 시프트를 수행하는 것으로 결정되는 것에 응답하여, 제1 비트패턴 또는 제2 비트패턴 중 적어도 하나에 대해 제1 시프트를 수행하는 단계, 제1 비트패턴 및 제2 비트패턴에 대해 제2 시프트를 수행하는 단계 및 비트 시프트가 수행된 제1 비트패턴 및 제2 비트패턴의 이진 연산을 수행하는 단계를 포함하고, 비트 시프트는, 제2 시프트를 포함하고, 제1 시프트를 수행하는 것으로 결정되는 것에 응답하여 제1 시프트를 포함한다. The present disclosure relates to a bit pattern operation method using dynamic bit shifts performed by an arithmetic unit. A bit pattern operation method using dynamic bit shifts comprises: a step of determining whether to perform a first shift for exponential alignment of a first bit pattern associated with a first operand and a second bit pattern associated with a second operand; a step of performing a first shift on at least one of the first bit pattern or the second bit pattern in response to the determination to perform the first shift; a step of performing a second shift on the first bit pattern and the second bit pattern; and a step of performing a binary operation on the first bit pattern and the second bit pattern on which the bit shifts have been performed. The bit shift includes the second shift and includes the first shift in response to the determination to perform the first shift.","assignee":"리벨리온 주식회사","inventors":["김진석"],"publication_date":"2026-04-16","filing_date":"2026-04-01","priority_date":"2023-09-22","cpc_codes":["G","G06","G06F","G06F5/00","G06F5/01","G06F5/012","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/483","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/483","G06F7/487","G06F7/4876","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/50","G06F7/501","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/52","G06F7/523","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/544","G06F7/5443","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/57","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260051318A/en"},{"publication_number":"AU2026202443A1","title":"Systems and methods for automated configuration to order and quote to order","abstract":"Computerized systems and methods are disclosed for automating Configure to Order (CTO) and Quote to Order (QTO) processes. Methods include receiving user inputs for desired product configurations, retrieving corresponding data from a bill of materials database, and calculating optimized pricing through intelligent rules based on real-time market data. Automated quotes are generated and transferred to orders in a vendor system, selected based on pre-set criteria like vendor reputation and delivery time. Validation steps reduce errors, and real-time reports are generated. The system integrates a Real-Time Data Mesh for data aggregation, a Single Pane of Glass User Interface for user interactions, and Advanced Analytics and Machine Learning Modules for implementing rule-based and learning algorithms. The system is accessible across various devices and standardizes data for uniform consumption, while also employing machine learning models to continually optimize processes. Notifications are sent to users upon successful execution of orders or completion of quotes. 20 26 20 24 43 31 M ar 2 02 6 A B S T R A C T 2 0 2 6 2 0 2 4 4 3 3 1 M a r 2 0 2 6","assignee":"Ingram Micro Inc","inventors":["Sanjib Sahoo"],"publication_date":"2026-04-16","filing_date":"2026-03-31","priority_date":"2024-02-21","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087","G06Q10/0875","G","G06","G06F","G06F16/00","G06F16/20","G06F16/23","G06F16/2379","G","G06","G06F","G06F18/00","G06F18/20","G06F18/27","G","G06","G06N","G06N20/00","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0206","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0283","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0621","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0633","G06Q30/0635"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202443A1/en"},{"publication_number":"AU2026202430A1","title":"Information extraction from daily drilling reports using machine learning","abstract":"22554957_1 (GHMatters) P118513.AU.1 A system and method are provided for extracting information regarding a drill site including forming one or more documents having one or more raw comments regarding a well site. Raw data may be extracted from the one or more documents to produce extracted raw data. The extracted raw date may be pre-processed by removing ambiguity, artifacts, and/or formatting errors from the one or more raw comments to produce pre-processed data. Topics data may be extracted from the pre-processed data using a natural language processing (NLP) algorithm to produce extracted topics data. Measurement data may also be extracted from the pre-processed data using the NLP algorithm to produce extracted measurement data. The extracted topics data and the extracted measurement data may be aggregated to form a set of discrete data points, such as calibration points, per comment to produce aggregated data and one more calibration points may be identified from the aggregated data. The results of the one or more calibration points may then be presented.","assignee":"Geoquest Systems BV","inventors":["Athithan DHARMARATNAM","Ivan DIAZ GRANADOS PERTUZ","Karsten Fischer","Francisco Jose GOMEZ","Mohamed Saad KISRA"],"publication_date":"2026-04-16","filing_date":"2026-03-31","priority_date":"2019-09-13","cpc_codes":["E","E21","E21B","E21B47/00","E21B47/12","E","E21","E21B","E21B47/00","E","E21","E21B","E21B41/00","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/08","E","E21","E21B","E21B2200/00","E21B2200/20","E","E21","E21B","E21B2200/00","E21B2200/22","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202430A1/en"},{"publication_number":"AU2026202439A1","title":"5G resource assignment technique","abstract":"A processor, comprising: one or more circuits to select one or more wireless devices to be assigned to a group associated with a frequency resource, the group being for transmitting signals to the one or more wireless devices, the assignment being based, at least in part, on parallel executions of a group selection algorithm performed by one or more 5 graphics processing units (GPUs) communicatively coupled to the processor.","assignee":"Nvidia Corp","inventors":["Harsha Deepak Banuli Nanje Gowda","James Hansen DELFELD","Yan Huang"],"publication_date":"2026-04-16","filing_date":"2026-03-31","priority_date":"2019-10-30","cpc_codes":["H","H04","H04W","H04W28/00","H04W28/16","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","H","H04","H04B","H04B7/00","H04B7/02","H04B7/04","H04B7/0413","H04B7/0452","H","H04","H04B","H04B7/00","H04B7/02","H04B7/04","H04B7/0413","H04B7/0456","H","H04","H04B","H04B7/00","H04B7/02","H04B7/04","H04B7/06","H04B7/0686","H04B7/0691","H","H04","H04W","H04W72/00","H04W72/04","H04W72/044","H04W72/0453","G","G06","G06N","G06N20/00"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202439A1/en"},{"publication_number":"KR20260051301A","title":"AI-driven Tagged Document Reconstruction System via User Edit Tracking and Retroactive Rule Application","abstract":"원시 문서(Original Document)와 상기 원시 문서에 기초하여 외부 도구에 의해 태그가 부여된 XML 기반 문서를 처리하는 인공지능(AI) 에이전트 시스템에 있어서, 상기 원시 문서와 상기 XML 기반 문서를 비교 분석하여, 문장 분할 단위 및 각 문장에 할당된 태그의 배치 특성을 포함하는 태그 할당 규칙을 추출하는 구조 분석부; 상기 원시 문서의 단락 구조를 유지하되 상기 태그 정보가 제거된 편집용 문서를 생성하는 문서 변환부; 상기 편집용 문서 상에서 수행된 사용자의 편집 내용을 모니터링하여, 수정된 텍스트가 상기 원시 문서 및 XML 기반 문서 내의 어느 구조적 위치 및 태그 범위에 대응하는지 매핑 정보를 생성하는 편집 추적부; 및 상기 편집 추적부로부터 제공된 매핑 정보에 기초하여, 상기 구조 분석부에서 파악된 태그 할당 규칙을 상기 수정된 텍스트에 역으로 적용함으로써, 상기 외부 도구의 포맷과 호환되는 태그가 삽입된 최종 문서를 재구성하는 결과 생성부;를 포함하는 것을 특징으로 하는 AI 에이전트 시스템이 제공된다. An AI agent system for processing an original document and an XML-based document tagged by an external tool based on the original document is provided, comprising: a structural analysis unit that compares and analyzes the original document and the XML-based document to extract tag assignment rules including sentence division units and placement characteristics of tags assigned to each sentence; a document conversion unit that generates an editable document in which the tag information is removed while maintaining the paragraph structure of the original document; an edit tracking unit that monitors user edits performed on the editable document to generate mapping information on which structural location and tag range within the original document and the XML-based document the modified text corresponds to; and a result generation unit that reconstructs a final document with inserted tags compatible with the format of the external tool by inversely applying the tag assignment rules identified by the structural analysis unit to the modified text based on the mapping information provided by the edit tracking unit.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-16","filing_date":"2026-03-30","priority_date":"2026-03-30","cpc_codes":["G","G06","G06F","G06F40/00","G06F40/10","G06F40/103","G06F40/117","G","G06","G06F","G06F40/00","G06F40/10","G06F40/166","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G","G06","G06F","G06F40/00","G06F40/20","G06F40/253","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q50/00","G06Q50/10"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260051301A/en"},{"publication_number":"KR20260049762A","title":"A method for operating a system for determining the re-commercialization of returned products","abstract":"반품 상품의 분류 시스템이 개시된다. 본 분류 시스템은, 반품 상품에 대한 복수의 점검 문항에 대한 체크리스트를 제공하고, 체크리스트에 대한 사용자 입력을 수신하는, 검수자 단말, 반품 상품의 외관 이미지를 획득하는 카메라 장치, 체크리스트에 대한 사용자 입력 및 외관 이미지를 바탕으로 반품 상품을 등급화 하는 인공지능 모델을 포함하는, 등급 판정 장치를 포함한다. A sorting system for returned goods is disclosed. The sorting system includes an inspector terminal that provides a checklist of multiple inspection items for returned goods and receives user input for the checklist; a camera device that acquires an external image of a returned product; and a grading device that includes an artificial intelligence model that grades the returned product based on the user input for the checklist and the external image.","assignee":"유상호","inventors":["유상호","김용준"],"publication_date":"2026-04-14","filing_date":"2026-04-06","priority_date":"2023-11-03","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06395","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G06Q10/06375","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/083","G06Q10/0837","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0278"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260049762A/en"},{"publication_number":"KR20260049511A","title":"Device and method for analyzing functional modalities","abstract":"실시예들은 기능적 분석 장치로서, 하나 이상의 모달리티로부터 정보를 획득하는 획득 모듈; 서로 다른 모달리티 사이 또는 단일 모달리티 내의 서로 다른 기본정보 사이의 연관정보를 생성하는 기능적 분석 모듈; 및 상기 기본정보와 연관된 하나 이상의 연관정보 및 통합정보 중 적어도 하나를 포함하는 기능적 분석 결과를 제공하는 분석 결과 제공 모듈;을 포함한다. 서로 다른 모달리티(modality) 또는 모달리티(modality) 간에 연관정보 및 통합정보를 제공함으로써 진단에 대한 정확도를 증가시킬 수 있고, 진단 세분화 및 처치 선택, 원인 판단을 도울 수 있으며, 복수의 모달리티에서 발견되는 위험한 소견을 확인하여 광범위하게 평가하고 빠른 응급 처치를 통해 환자 안전 보장을 향상할 수 있다. The embodiments include a functional analysis device comprising: an acquisition module that acquires information from one or more modalities; a functional analysis module that generates association information between different modalities or between different basic information within a single modality; and an analysis result providing module that provides a functional analysis result including at least one of one or more association information and integrated information associated with the basic information. By providing association information and integrated information between different modalities or modalities, the accuracy of diagnosis can be increased, and the diagnostic refinement, treatment selection, and cause determination can be aided. Additionally, dangerous findings found in multiple modalities can be identified and evaluated extensively, and patient safety can be enhanced through rapid emergency treatment.","assignee":"서울대학교병원","inventors":["김중희"],"publication_date":"2026-04-14","filing_date":"2026-04-03","priority_date":"2022-06-07","cpc_codes":["A","A61","A61B","A61B6/00","A61B6/52","A61B6/5211","A61B6/5217","A","A61","A61B","A61B5/00","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/318","A61B5/33","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/318","A61B5/346","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7275","A","A61","A61B","A61B6/00","A","A61","A61B","A61B6/00","A61B6/52","A61B6/5294","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G16","G16H","G16H30/00","G16H30/20","G","G16","G16H","G16H30/00","G16H30/40","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H70/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260049511A/en"},{"publication_number":"KR20260049755A","title":"System for managing safety linked to cognitive information based image","abstract":"작업현장에서 기존의 영상 모니터링에 활용되는 일반 CCTV(Closed Circuit Television) 시스템만을 활용하여 추가 감지센서 및 장비 사용 없이 CCTV의 촬영 영상에 대한 영상인식을 통하여 위험 및 주의 상황을 인지할 수 있는 영상기반 인지정보를 연계한 안전 관리 시스템이 개시된다. 상기 안전 관리 시스템은, 작업장의 영상 정보를 수집하는 영상 수집부, 수집된 상기 영상 정보에 대하여 기계 학습을 통해 영상잡음을 분류하여 상황 인지 정보를 생성하고 상기 영상잡음을 제거하여 잡음 제거 영상 정보를 생성하는 영상 처리부, 및 상기 상황 인지 정보의 실시간 모니터링을 통해 상기 작업장의 환경 농도를 판단하여 상기 작업장의 조명 또는 환기 설비를 자동으로 제어하는 제어 관리부를 포함하는 것을 특징으로 한다. A safety management system linked with image-based recognition information is disclosed, which can recognize dangerous and cautionary situations through image recognition of captured CCTV footage without using additional detection sensors or equipment, by utilizing only a standard CCTV (Closed Circuit Television) system used for existing video monitoring at a work site. The safety management system is characterized by comprising: an image collection unit that collects image information of a workplace; an image processing unit that generates situational recognition information by classifying image noise through machine learning on the collected image information and generates noise-removed image information by removing the image noise; and a control management unit that determines the environmental concentration of the workplace through real-time monitoring of the situational recognition information and automatically controls the lighting or ventilation facilities of the workplace.","assignee":"한국전력공사","inventors":["박준범","박윤식","박예슬","김대현","김영준"],"publication_date":"2026-04-14","filing_date":"2026-04-02","priority_date":"2021-11-25","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06F","G06F18/00","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06T","G06T5/00","G06T5/70","G","G06","G06T","G06T7/00","G06T7/10","G06T7/13","H","H04","H04N","H04N7/00","H04N7/18","G","G06","G06F","G06F2218/00","G06F2218/02","G06F2218/04"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260049755A/en"},{"publication_number":"KR20260050348A","title":"Wafer-scale intelligent computing device","abstract":"본 발명은 단일 실리콘 웨이퍼 상에 형성된 복수의 다이를, 다이들 사이의 스크라이브 라인 영역 상에 형성된 다이 간 인터커넥트 배선층으로 직접 연결하여 하나의 거대한 연산 패브릭으로 통합하는 웨이퍼 스케일 지능형 연산 소자에 관한 것이다. 각 다이는 복수의 독립적인 연산 코어와 코어별 전용 로컬 SRAM을 포함하는 MIMD 구조의 분산형 병렬 AI 연산 가속기 아키텍처로 구성된다. MIMD 구조 특유의 연산-데이터 공존 고장 문제를 해결하기 위하여, 로컬 SRAM 체크포인트 미러링(청구항 2), MIMD 실행 상태 스냅샷 전파(청구항 3), 그래프 파티션 재할당 프로토콜(청구항 4), 결함 밀도 맵 기반 선제적 파티션 배치(청구항 5)의 4계층 결함 허용 메커니즘을 구비하며, 다이 간 인터커넥트 배선층 내 제1 및 제2 전용 채널을 통해 데이터 보존과 실행 상태 보존을 동시 수행하는 구조(청구항 6)를 포함한다. The present invention relates to a wafer-scale intelligent computing device that integrates a plurality of dies formed on a single silicon wafer into one large computing fabric by directly connecting them to an inter-die interconnect wiring layer formed on a scribe line region between the dies. Each die is configured as a distributed parallel AI computing accelerator architecture with a MIMD structure, comprising a plurality of independent computing cores and a dedicated local SRAM per core. To solve the computation-data coexistence failure problem unique to the MIMD structure, the device is equipped with a four-layer fault-tolerant mechanism comprising local SRAM checkpoint mirroring (Claim 2), MIMD execution state snapshot propagation (Claim 3), a graph partition reallocation protocol (Claim 4), and a fault density map-based preemptive partition placement (Claim 5), and includes a structure (Claim 6) that simultaneously performs data preservation and execution state preservation through first and second dedicated channels within the inter-die interconnect wiring layer.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-14","filing_date":"2026-03-28","priority_date":"2026-03-28","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260050348A/en"},{"publication_number":"KR20260050347A","title":"A Fault Recovery Method and System for Static Scheduling-Based Multi-Chip Inference Systems Utilizing Partial Rescheduling and Deterministic Elasticity","abstract":"본 발명은 컴파일 타임에 확정된 정적 실행 스케줄(140)을 기반으로 동작하는 멀티칩 추론 시스템(100)에서 추론 처리 유닛(110)의 장애 발생 시, 전체 재컴파일 없이 수십 마이크로초 이내에 복구를 완료하는 방법 및 시스템을 제공한다. 본 발명은 장애 유닛의 칩별 스케줄 세그먼트(141)를 잔존 유닛들의 가용 온칩 메모리(111)에 재배치하는 공간 기반 재파티셔닝과, 정적 실행 스케줄(140) 내의 슬랙 타임(142)에 장애 태스크를 할당하는 사전 생성 델타 스케줄(143) 기반 명령어 패칭을 이중 전략으로 제공한다. 가변 지연 버퍼(114)는 복구 이후에도 결정론적 출력 타이밍을 보장함으로써, 자율주행 및 의료 기기 등 안전 필수 응용 분야에서의 실시간 신뢰성을 확보한다. The present invention provides a method and system for completing recovery within tens of microseconds without full recompilation when a failure occurs in an inference processing unit (110) in a multichip inference system (100) that operates based on a static execution schedule (140) determined at compile time. The present invention provides a dual strategy of space-based repartitioning, which relocates the chip-specific schedule segment (141) of the failed unit to the available on-chip memory (111) of the remaining units, and pre-generated delta schedule (143)-based instruction patching, which assigns the failed task to a slack time (142) within the static execution schedule (140). A variable delay buffer (114) ensures deterministic output timing even after recovery, thereby securing real-time reliability in safety-critical applications such as autonomous driving and medical devices.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-14","filing_date":"2026-03-28","priority_date":"2026-03-28","cpc_codes":["G","G06","G06F","G06F11/00","G06F11/07","G06F11/0703","G06F11/0793","G","G06","G06F","G06F11/00","G06F11/07","G06F11/16","G06F11/1675","G","G06","G06F","G06F11/00","G06F11/07","G06F11/16","G06F11/20","G06F11/202","G06F11/2023","G06F11/2028","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/48","G06F9/4806","G06F9/4843","G06F9/4881","G","G06","G06N","G06N5/00","G06N5/04"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260050347A/en"},{"publication_number":"KR20260049488A","title":"Adaptive Tier Classification and Quantization System and Method Based on Layer Characteristics of Artificial Intelligence Neural Networks","abstract":"본 발명은 인공지능(AI) 신경망 모델의 연산 효율을 최적화하는 기술에 관한 것으로, 보다 상세하게는신경망을 구성하는 각 중간층(Intermediate Layer)의 구조적 속성(Structural Attribute) 또는 연산 특성(Operational Characteristic)에 맞추어 데이터를 동적으로 티어(Tier)별로 분류하고, 티어별로 차등적인 비트 정밀도(Bit Precision)를 적용하여 양자화(Quantization)를 수행하는 시스템 및 방법에 관한 것이다. 특히 본 발명은 합성곱 층(Convolutional Layer), 어텐션 층(Attention Layer), 정규화 층(Normalization Layer), 임베딩 층(Embedding Layer), 완전연결 층(Fully-Connected Layer) 등 이종(異種) 층으로 구성된 신경망 모델에서 각 층의 데이터 특성에 최적화된 티어 분류 알고리즘을 개별적으로 할당하고 동적으로 교체하는 적응형 정책 관리 기술에 관한 것이다. The present invention relates to a technology for optimizing the computational efficiency of an artificial intelligence (AI) neural network model, and more specifically, to a system and method for dynamically classifying data by tier according to the structural attribute or operational characteristic of each intermediate layer constituting the neural network, and performing quantization by applying differential bit precision to each tier. In particular, the present invention relates to an adaptive policy management technology for individually assigning and dynamically replacing a tier classification algorithm optimized for the data characteristics of each layer in a neural network model composed of heterogeneous layers such as a convolutional layer, an attention layer, a normalization layer, an embedding layer, and a fully-connected layer.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-14","filing_date":"2026-03-26","priority_date":"2026-03-26","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260049488A/en"},{"publication_number":"KR20260049145A","title":"AI Persona Generation System and Method Using Multidimensional Interaction Mapping and Closed-Loop Consistency Verification Based on Big Five Personality Traits","abstract":"본 발명은 Big Five 성격 모델의 개방성(O), 성실성(C), 외향성(E), 우호성(A), 신경증(N) 5가지 특성 각각에 대한 0~100 연속 수치를 입력받아, 각 특성 수치가 응답 생성 온도, 어휘 도메인 가중치, 감성 표현 수준, 문장 구조 선호도의2개 이상 차원에 동시에 교차 기여하는 다차원 상호작용 매핑을 수행하는 파라미터 매핑부와, 생성된 응답의 성격 적합도를 실시간으로 측정하여 일관성 오차를 파라미터 매핑부에 피드백 보정하는 클로즈드 루프 구조의 페르소나 생성부를 결합함으로써, 목표 성격 특성과 일관된 AI 페르소나를 재현 가능하게 생성하는 시스템 및 방법을 제공한다. The present invention provides a system and method for reproducibly generating an AI persona consistent with target personality traits by combining a parameter mapping unit that receives a continuous numerical value from 0 to 100 for each of the five traits of the Big Five personality model—openness (O), conscientiousness (C), extraversion (E), agreeableness (A), and neuroticism (N)—and performs a multidimensional interaction mapping in which each trait value simultaneously cross-contributes to two or more dimensions, such as response generation temperature, vocabulary domain weight, emotional expression level, and sentence structure preference, and a persona generation unit with a closed-loop structure that measures the personality fit of the generated response in real time and feeds back the consistency error to the parameter mapping unit.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-13","filing_date":"2026-03-25","priority_date":"2026-03-25","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/3332","G06F16/3334","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/335","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F16/00","G06F16/30","G06F16/34","G06F16/345","G","G06","G06F","G06F40/00","G06F40/30","G06F40/35","G","G06","G06N","G06N20/00","G","G06","G06F","G06F2201/00","G06F2201/81"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260049145A/en"},{"publication_number":"KR20260049150A","title":"Operation server for ai automated investment analysis based on aviation asset data and operation method thereof","abstract":"항공자산 데이터 기반 AI 자동화 투자 분석을 위한 운영 서버가 개시된다. 상기 운영 서버는, 적어도 하나의 항공 서버 또는 관리자의 입력을 통해 항공 자산 데이터를 획득하고, 획득된 상기 항공 자산 데이터를 이용하여 핵심 자산 정보를 생성하는 항공 자산 데이터 수집부; 상기 핵심 자산 정보를 이용하여 인공지능 기반으로 항공 자산에 대한 미래 시점의 매도가격을 예측하는 인공지능 가격 예측부; 예측된 상기 매도가격을 기반으로 상기 항공 자산에 대한 투자 수익성 분석을 수행하는 투자 분석부; 상기 항공 자산에 대한 시장성 분석을 수행하는 시장 분석부; 및 상기 투자 수익성 분석과 상기 시장성 분석을 기반으로 투자 리포트를 생성하는 투자 리포트 생성부를 포함한다. An operating server for AI automated investment analysis based on aviation asset data is disclosed. The operating server comprises: an aviation asset data collection unit that acquires aviation asset data through input from at least one aviation server or manager and generates core asset information using the acquired aviation asset data; an AI price prediction unit that predicts a future selling price of an aviation asset based on artificial intelligence using the core asset information; an investment analysis unit that performs an investment profitability analysis of the aviation asset based on the predicted selling price; a market analysis unit that performs a marketability analysis of the aviation asset; and an investment report generation unit that generates an investment report based on the investment profitability analysis and the marketability analysis.","assignee":"브이엠아이씨주식회사","inventors":["남교훈","김봉선"],"publication_date":"2026-04-13","filing_date":"2026-03-25","priority_date":"2025-06-20","cpc_codes":["G","G06","G06Q","G06Q40/00","G06Q40/06","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06Q","G06Q10/00","G06Q10/10","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0278","G","G06","G06V","G06V30/00","G06V30/10"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260049150A/en"},{"publication_number":"KR20260049146A","title":"AI Agent System and Method Specialized for Word Processing, Featuring Automated Data Substitution Based on User Task Pattern Learning across Task Categories","abstract":"본 발명은 워드 프로세서(200) 상에서 사용자가 수행하는 '찾기 및 바꾸기' 동작을 이벤트 모니터링 모듈(110)이 실시간으로 감지하여 변경 전 데이터(Source) 및 변경 후 데이터(Target)로 이루어진 데이터 쌍을 추출하고, 데이터매핑 및 분류 엔진(120)이 문서 메타데이터와 NLP 기반 키워드 분석을 통해 해당 데이터 쌍을 작업 카테고리별로 분류하며, 작업별 데이터 테이블 관리부(130)가 카테고리별 데이터 테이블(131) 및 공용 데이터 테이블(132)에 구조화하여 저장하고, 자동 실행 제어 모듈(140)이 동일 카테고리의 신규 문서 작업 이벤트 발생 시 저장된 치환 규칙을 일괄 적용하여 자동 실행 결과물을 생성하는, 사용자 작업 패턴 학습 기반의 작업 카테고리별 데이터 치환 자동화 기능을 구비한 워드 프로세싱 특화 AI 에이전트 시스템(100) 및 방법에 관한 것이다. 본 발명에 의하면, 반복적 수동 치환 작업이 자동화되어 생산성이 향상되고, 도메인별 일관된 용어 관리가 가능하며, 사용자의 작업 패턴이 축적될수록 치환 규칙 베이스가 자동 최적화되는 자기 진화형 에이전트를 실현할 수 있다. The present invention relates to a word processing specialized AI agent system (100) and a method equipped with a task category-specific data replacement automation function based on user task pattern learning, wherein an event monitoring module (110) detects in real time a 'find and replace' operation performed by a user on a word processor (200) and extracts a data pair consisting of data before change (Source) and data after change (Target); a data mapping and classification engine (120) classifies the data pair by task category through document metadata and NLP-based keyword analysis; a task-specific data table management unit (130) structures and stores the data in a category-specific data table (131) and a common data table (132); and an automatic execution control module (140) generates an automatic execution result by applying the stored replacement rules in bulk when a new document task event of the same category occurs. According to the present invention, repetitive manual replacement work is automated to improve productivity, consistent term management by domain is possible, and a self-evolving agent can be realized in which the replacement rule base is automatically optimized as the user's task patterns accumulate.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-13","filing_date":"2026-03-25","priority_date":"2026-03-25","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06F","G06F16/00","G06F16/20","G","G06","G06F","G06F40/00","G06F40/10","G06F40/166","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G","G06","G06F","G06F40/00","G06F40/20","G06F40/253","G","G06","G06F","G06F40/00","G06F40/40","G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260049146A/en"},{"publication_number":"KR20260049123A","title":"A building control method controlling building by analyzing difference graph using a deep learning algorithm and a building control system performing the same","abstract":"딥러닝 알고리즘을 활용한 괴리 그래프 분석을 통해 빌딩을 제어하는 빌딩 제어 방법 및 이를 수행하는 빌딩 제어 시스템에 관한 것이다. 본 개시의 기술적 사상에 따른 적어도 하나의 프로세서에 의해 수행되는, 대상 빌딩을 제어하기 위한 제어 출력 값을 결정하는 빌딩 제어 방법은, 복수의 시점 별 입력 값 및 목표 값을 수신하는 단계; 상기 입력 값과 목표 값의 차이에 대응하는 괴리 값에 관한 괴리 그래프를 생성하는 단계; 상기 괴리 그래프를 분석함으로써 시간 상수를 결정하는 단계; 딥러닝 알고리즘을 활용하여 상기 대상 빌딩에 대응하는 비례 상수를 결정하는 단계; 상기 시간 상수 및 상기 비례 상수를 활용하여 제어 신호 모델을 결정하는 단계;및 상기 제어 신호 모델을 통해 생성한 제어 출력 값을 이용하여 상기 대상 빌딩을 제어하는 단계;를 포함할 수 있다. The present invention relates to a building control method for controlling a building through divergence graph analysis using a deep learning algorithm, and a building control system for performing the same. A building control method for determining a control output value for controlling a target building, performed by at least one processor according to the technical concept of the present disclosure, may include: receiving input values and target values at a plurality of time points; generating a divergence graph regarding divergence values corresponding to the difference between the input values and the target values; determining a time constant by analyzing the divergence graph; determining a proportionality constant corresponding to the target building using a deep learning algorithm; determining a control signal model using the time constant and the proportionality constant; and controlling the target building using a control output value generated through the control signal model.","assignee":"주식회사 동양이엔씨","inventors":["최두현"],"publication_date":"2026-04-13","filing_date":"2026-01-29","priority_date":"2026-01-29","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06F","G06F17/00","G06F17/10","G06F17/11","G06F17/13","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","H","H04","H04L","H04L67/00","H04L67/01","H04L67/12","H04L67/125"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260049123A/en"},{"publication_number":"KR102951791B1","title":"Server, system, method and program providing location identification and mapping services included in artificial intelligence-based posting content","abstract":"본 발명의 실시 예에 따르면, 장소 식별 및 매핑 서비스를 제공하는 서버가 제공된다. 상기 서버는, SNS 서버로부터 포스팅 정보를 수신하고, 상기 포스팅 정보에 포함된 포스팅 정보를 LLM 에이전트에 제공하여, 장소명, 지역 및 주소를 포함하는 장소 정보를 생성하는 장소정보 생성모듈; 장소 정보에 포함된 주소를 이용하여 지도 서비스 서버로부터 주소와 매칭되고 위도 좌표, 경도 좌표 및 주소를 포함하는 좌표 정보를 수신하고, 주소 정보와 매칭되는 좌표 정보를 결정하는 좌표 결정모듈; 및 상기 좌표 결정모듈에 의해 결정된 좌표 정보 및 장소 정보를 이용해 장소 정보와 매칭되는 최종 장소를 결정하는 장소 결정모듈을 포함하고, 상기 장소 결정모듈은, 상기 지도 서비스 서버로부터 결정된 좌표 정보와 매칭되는 복수의 장소 후보를 수신하고, 알고리즘 기반의 문자열 매칭 및 LLM 에이전트를 활용한 매칭 중 적어도 하나를 이용해 복수의 장소 후보 중 장소 정보와 매칭되는 최종 장소를 결정한다. According to an embodiment of the present invention, a server providing a place identification and mapping service is provided. The server comprises: a place information generation module that receives posting information from an SNS server and provides the posting information included in the posting information to an LLM agent to generate place information including a place name, region, and address; a coordinate determination module that receives coordinate information from a map service server that matches the address and includes latitude coordinates, longitude coordinates, and an address, using the address included in the place information, and determines coordinate information that matches the address information; and a place determination module that determines a final place that matches the place information using the coordinate information and place information determined by the coordinate determination module. The place determination module receives a plurality of place candidates that match the determined coordinate information from the map service server, and determines a final place that matches the place information among the plurality of place candidates using at least one of algorithm-based string matching and matching using an LLM agent.","assignee":"(주)핫플","inventors":["조성진"],"publication_date":"2026-04-13","filing_date":"2025-11-27","priority_date":"2025-11-27","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/40","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G06F16/9537","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102951791B1/en"},{"publication_number":"KR102951337B1","title":"Real-time issue generation and evolution analysis and industry/asset impact prediction system","abstract":"본 발명은 실시간 이슈 생성·진화 분석 및 연관 산업·자산 영향도 예측 시스템에 관한 것으로서, 인공신경망을 이용한 실시간 이슈 생성·진화 분석 및 연관 산업·자산 영향도 예측 시스템에 있어서, 다중매체로부터 실시간으로 발생하는 이슈에 대한 텍스트, 이미지, 영상, 오디오 데이터 중 적어도 하나를 포함하는 데이터를 수집하는 데이터 수집부; 상기 데이터 수집부에서 수집된 데이터를 기반으로, 내용 분석 정보와 출처 분석 정보로 가공하고 이를 통합하여 이슈 메타데이터를 생성하는 데이터 가공부; 상기 데이터 가공부에서 생성된 상기 이슈 메타데이터를 기반으로, 신뢰성 점수와 화제성 점수를 산출하고, 이를 기반으로 이슈 중요도를 산출하는 중요도 산출부; 상기 이슈 메타데이터와 상기 이슈 중요도를 기반으로 관련 산업 또는 자산군의 예상 영향도를 산출하고 경제적 가치의 변화 방향성 및 정도를 예측하는 영향도 계산부; 및 상기 영향도 계산부의 예측 결과를 사용자 단말기에 제공하는 결과 제공부;를 포함한다. 본 발명에 따르면, 사용자는 실시간으로 발생하는 이슈에 대하여 산업 또는 자산군에 미치는 영향을 신속하게 파악할 수 있고, 따라서 금융 투자 의사결정자가 빠른 대응전략을 수립할 수 있다. The present invention relates to a real-time issue generation and evolution analysis and related industry and asset impact prediction system. The real-time issue generation and evolution analysis and related industry and asset impact prediction system using an artificial neural network comprises: a data collection unit that collects data including at least one of text, image, video, and audio data regarding issues occurring in real time from multimedia; a data processing unit that processes the data collected by the data collection unit into content analysis information and source analysis information, integrates them, and generates issue metadata; an importance calculation unit that calculates a reliability score and a topicality score based on the issue metadata generated by the data processing unit, and calculates the issue importance based thereon; an impact calculation unit that calculates the expected impact of a related industry or asset group and predicts the direction and extent of change in economic value based on the issue metadata and the issue importance; and a result providing unit that provides the prediction result of the impact calculation unit to a user terminal. According to the present invention, users can quickly identify the impact of real-time issues on an industry or asset class, and thus financial investment decision-makers can establish rapid response strategies.","assignee":"뉴로퓨전 주식회사","inventors":["김현식","조성민"],"publication_date":"2026-04-13","filing_date":"2025-11-26","priority_date":"2025-11-26","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G06Q10/06375","G","G06","G06F","G06F16/00","G06F16/90","G06F16/907","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/951","G","G06","G06F","G06F40/00","G06F40/20","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06Q","G06Q10/00","G06Q10/04"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102951337B1/en"},{"publication_number":"KR102951878B1","title":"A Risk Management System for Monitoring and Prediction of Hazardous Chemical Substance Risk Situations utilizing Machine Vision","abstract":"본 발명은 유해화학물질 취급 현장의 위험을 관리하는 자가 진단 및 자율 대응 시스템에 관한 것이다. 본 발명의 위험관리시스템은 Wi-SUN Mesh 네트워크 기반 지능형 센서 네트워크부, 영상 데이터 수집부, 중앙 처리 서버를 포함한다. 중앙 처리 서버는 IoT 센서의 시계열 데이터, CCTV 영상의 공간적 특징, 네트워크 상태 데이터를 융합하여 확장된 상태 벡터를 생성하는 다중 소스 데이터 융합부와, 디지털 트윈 환경에서 강화학습을 통해 누적 위험 회피 보상을 최대화하는 최적 대응 정책을 자율 학습하는 강화학습 기반 자율 대응 결정부, 및 실시간 컨텍스트 정보를 반영하여 동적 표준 운영 절차(SOP)를 생성·전파하는 전자 SOP 실행 및 상황 전파부를 포함한다. 본 발명은 위험 예측 기반 선제적 통신 관리를 통해 골든타임 내 안정적 대응을 보장하며, 사전 정의된 시나리오에 의존하는 종래 기술의 한계를 극복하여 복잡한 재난 상황에서도 진정한 자율 대응을 가능하게 한다. The present invention relates to a self-diagnosis and autonomous response system for managing risks at hazardous chemical handling sites. The risk management system of the present invention includes an intelligent sensor network unit based on a Wi-SUN Mesh network, a video data collection unit, and a central processing server. The central processing server includes a multi-source data fusion unit that generates an extended state vector by fusing time-series data from IoT sensors, spatial features of CCTV video, and network state data; a reinforcement learning-based autonomous response decision unit that autonomously learns an optimal response policy to maximize accumulated risk avoidance rewards through reinforcement learning in a digital twin environment; and an electronic SOP execution and situation dissemination unit that generates and propagates dynamic standard operating procedures (SOPs) by reflecting real-time context information. The present invention ensures a stable response within the golden time through risk prediction-based preemptive communication management, and enables true autonomous response even in complex disaster situations by overcoming the limitations of conventional technology that relies on predefined scenarios.","assignee":"주식회사 주빅스","inventors":["정계명","손도선"],"publication_date":"2026-04-13","filing_date":"2025-10-28","priority_date":"2025-10-28","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G16","G16Y","G16Y40/00","G16Y40/10","H","H04","H04N","H04N7/00","H04N7/18"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102951878B1/en"},{"publication_number":"LU603446B1","title":"Method, system, and device for intelligent refrigeration system efficiency optimization based on deep learning","abstract":"The present invention provides a method, system, and device for intelligent refrigeration system efficiency optimization based on deep learning, falling within the technical field of intelligent refrigeration system efficiency optimization. Firstly, it collects historical operating parameters and environmental data of the refrigeration system, and integrates them into a sample set according to time stamps after preprocessing. Based on a hybrid model of convolutional neural network (CNN) and long short-term memory (LSTM) network, a state prediction model is constructed to predict the system’s operating state. Combined with the state prediction model’s prediction of data at the next future moment, an optimization objective function is defined, and a deep Q-network (DQN) model is used to formulate the optimal adjustment strategy for system operation. Finally, it monitors the current operating parameters and environmental data in real time, calculates a comprehensive safety score, and compares the score with a safety threshold to decide whether to continue efficiency optimization or give priority to troubleshooting. This method realizes efficient and safe operation optimization of the refrigeration system by combining deep learning with physical modeling.","assignee":"Univ Wuxi","inventors":["Chenghao Yu","Yuan Ding","Tianyu Xie","Zhengcheng Wang","Daiting Zheng","Chuxin Fu","Moyi Dai","Qi Zhong","Shengyi Liu","Ning Sun","Weimin Mao","Jinying Liu","Yanxue Chen"],"publication_date":"2026-04-13","filing_date":"2025-10-13","priority_date":"2025-10-13","cpc_codes":["F","F25","F25B","F25B49/00","F25B49/02","F","F24","F24F","F24F1/00","G","G06","G06N","G06N3/00","F","F25","F25B","F25B2700/00","F25B2700/02","F","F25","F25B","F25B2700/00","F25B2700/13","F25B2700/133","F25B2700/1331","F","F25","F25B","F25B2700/00","F25B2700/15","F25B2700/151","F","F25","F25B","F25B2700/00","F25B2700/19","F25B2700/195","F","F25","F25B","F25B2700/00","F25B2700/19","F25B2700/197","F","F25","F25B","F25B2700/00","F25B2700/21","F25B2700/2116","F25B2700/21161","F","F25","F25B","F25B2700/00","F25B2700/21","F25B2700/2116","F25B2700/21163","F","F25","F25B","F25B2700/00","F25B2700/21","F25B2700/2117","F25B2700/21174"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU603446B1/en"},{"publication_number":"KR20260049107A","title":"System and method for ai/ml measurement reporting and handover triggering","abstract":"무선 통신 시스템에서 사용자 단말(ue)에 의해 수행되는 방법이 제공된다. 방법은, 서비스 셀(serving cell) 및 하나 이상의 후보 셀(candidate cell)에 관련된 측정값을 획득하는 단계, 측정값에 기초하여 입력 데이터를 생성하는 단계, 예측 시간 오프셋(prediction time advance, t)을 이용하여 인공지능(ai) 또는 기계학습(ml) 모델을 상기 입력 데이터에 적용하여, 측정 보고(measurement report)를 전송할지 또는 서비스 셀로부터 타깃 셀(target cell)로의 핸드오버를 개시할지를 결정하는 단계, 및 결정에 따라, 측정 보고를 전송하거나 핸드오버를 개시하는 단계 중 적어도 하나를 수행하는 단계를 포함한다. A method is provided to be performed by a user terminal (ue) in a wireless communication system. The method comprises the steps of: obtaining a measurement value related to a serving cell and one or more candidate cells; generating input data based on the measurement value; applying an artificial intelligence (AI) or machine learning (ML) model to the input data using a prediction time advance (t) to determine whether to transmit a measurement report or initiate a handover from the serving cell to a target cell; and performing at least one of the steps of transmitting a measurement report or initiating a handover according to the determination.","assignee":"삼성전자주식회사","inventors":["올람 압바스 칼릴리","배정현"],"publication_date":"2026-04-13","filing_date":"2025-10-10","priority_date":"2024-10-04","cpc_codes":["H","H04","H04W","H04W36/00","H04W36/0005","H04W36/0055","H04W36/0058","G","G06","G06N","G06N20/00","H","H04","H04B","H04B17/00","H04B17/30","H04B17/309","H04B17/318","H04B17/328","H","H04","H04L","H04L41/00","H04L41/16","H","H04","H04W","H04W36/00","H04W36/0005","H04W36/0083","H04W36/00837","H","H04","H04W","H04W36/00","H04W36/0005","H04W36/0083","H04W36/0085"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260049107A/en"},{"publication_number":"KR20260048532A","title":"A computing device for diagnosing a quality of data","abstract":"본 개시의 일 실시예에 따르면, 데이터 셋을 획득하여 상기 데이터 셋에 대한 진단 결과를 제공하기 위한 컴퓨팅 장치로서, 출력 장치; 메모리; 및 상기 메모리에 저장된 적어도 하나의 인스트럭션을 기초로 동작하는 적어도 하나의 프로세서;를 포함하고, 상기 적어도 하나의 프로세서는, 데이터 셋을 획득하고, 상기 데이터 셋을 잠재 공간에 매핑함으로써 제1 매니폴드를 획득하고- 이때, 상기 제1 매니폴드는 상기 데이터 셋에 대응되는 포인트 데이터 셋을 포함함-, 상기 포인트 데이터 셋에 포함된 포인트 데이터들 중 적어도 일부를 이미징 공간에 나타냄으로써 데이터 이미지를 획득하고, 상기 데이터 이미지 및 상기 데이터 이미지를 분석하여 획득된 추가 정보를 포함하는 진단 레포트를 상기 출력 장치를 통해 출력하는 컴퓨팅 장치가 제공될 수 있다. According to one embodiment of the present disclosure, a computing device for acquiring a data set and providing a diagnostic result for said data set comprises: an output device; a memory; and at least one processor operating based on at least one instruction stored in said memory; wherein the at least one processor acquires a data set and acquires a first manifold by mapping said data set into a latent space—wherein the first manifold includes a point data set corresponding to said data set—acquires a data image by displaying at least some of the point data included in said point data set into an imaging space, and outputs a diagnostic report including said data image and additional information acquired by analyzing said data image through said output device.","assignee":"주식회사 페블러스","inventors":["이주행","이정원"],"publication_date":"2026-04-10","filing_date":"2026-04-06","priority_date":"2022-06-29","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G06F16/358","G","G06","G06F","G06F16/00","G06F16/20","G06F16/26","G","G06","G06F","G06F16/00","G06F16/30","G06F16/34","G06F16/345","G","G06","G06F","G06F16/00","G06F16/50","G06F16/55","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260048532A/en"},{"publication_number":"KR20260048531A","title":"System for generating automatically question based on neural network by query generation","abstract":"서로 다르게 사전학습 및 미세 조정된 다수의 모델을 활용한 신경망 기반 질의 자동 생성 시스템이 개시된다. 상기 시스템은, 다수의 질의 생성 신경망을 기반으로 질의를 생성하고 생성된 상기 질의에 대한 점수를 생성하는 질의 생성 수행부, 및 생성된 상기 점수를 이용하여 최종 질의를 선정하는 질의 선정 수행부를 포함하는 것을 특징으로 한다. A neural network-based automatic query generation system utilizing multiple models that have been pre-trained and fine-tuned differently is disclosed. The system is characterized by comprising a query generation unit that generates a query based on multiple query generation neural networks and generates a score for the generated query, and a query selection unit that selects a final query using the generated score.","assignee":"한국전력공사","inventors":["황명하","신지강","임정선","서호진","조희"],"publication_date":"2026-04-10","filing_date":"2026-04-03","priority_date":"2022-06-21","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G","G06","G06F","G06F40/00","G06F40/20","G06F40/237","G06F40/242","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","H","H04","H04L","H04L51/00","H04L51/02"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260048531A/en"},{"publication_number":"KR20260048248A","title":"Method and system for providing artificial intelligence model containing multiple models","abstract":"본 개시의 일 실시예에 따른 방법은, 컴퓨터에 의해 실행되는 방법에 있어서, 상기 컴퓨터의 적어도 하나의 프로세서가, 복수의 전문 모델 및 라우터가 결합된 MoE(Mixture of Experts) 아키텍처 기반 모델에 포함된 상기 복수의 전문 모델 각각의 역할 및 기능을 특정하는 전문 모델 특성정보를 적어도 하나의 인공지능 모델을 통해 획득하는 단계; - 여기서, 상기 인공지능 모델은, 상기 복수의 전문 모델 및 라우터가 결합된 MoE 아키텍처 기반 모델에 대한 학습 수행 시 상기 복수의 전문 모델 각각에 대한 상기 라우터의 태스크 할당 상태를 추적하고, 상기 추적된 태스크 할당 상태에 기초하여 상기 각각의 전문 모델의 전문성을 판단함으로써 상기 전문 모델 특성정보를 생성하는 적어도 하나의 모델 특정화 모듈을 포함하고, 상기 복수의 전문 모델은, 상기 컴퓨터와 네트워크를 통해 연결된 외부 서버에 의해 제공되는 적어도 하나의 외부 모델을 포함하는 것을 특징으로 하고, 상기 적어도 하나의 프로세서가, 상기 획득된 전문 모델 특성정보를 이에 대응하는 전문 모델에 매칭하여 독립적으로 분리된 복수의 전문 모듈모델을 생성하고, 상기 생성된 복수의 전문 모듈모델을 데이터베이스화하여 저장하는 단계; 상기 적어도 하나의 프로세서가, 사용자 단말로부터 수신된 입력 데이터를 분석하여 상기 입력 데이터에 대응하는 도메인 정보를 획득하는 단계; 상기 적어도 하나의 프로세서가, 상기 획득된 도메인 정보 및 상기 데이터베이스화된 복수의 전문 모델 특성정보에 기초하여, 상기 도메인 정보에 대응하는 특성을 가지는 적어도 하나의 전문 모듈모델을 도메인 특화 전문모델로 결정하는 단계; 상기 적어도 하나의 프로세서가, 상기 결정된 적어도 하나의 도메인 특화 전문모델 및 상기 라우터를 결합하여 도메인 특화 MoE 모델을 구축하는 단계; 및 상기 적어도 하나의 프로세서가, 상기 구축된 도메인 특화 MoE 모델을 이용하여 상기 입력 데이터에 대한 출력 데이터를 생성하고 제공하는 단계를 포함한다. A method according to one embodiment of the present disclosure is a method executed by a computer, wherein at least one processor of the computer obtains expert model characteristic information that specifies the role and function of each of the plurality of expert models included in a Mixture of Experts (MoE) architecture-based model combined with a plurality of expert models and a router, through at least one artificial intelligence model; - wherein the artificial intelligence model includes at least one model specification module that generates the expert model characteristic information by tracking the task assignment status of the router for each of the plurality of expert models during the learning of the MoE architecture-based model combined with the plurality of expert models and a router, and determining the expertise of each of the expert models based on the tracked task assignment status, and wherein the plurality of expert models include at least one external model provided by an external server connected to the computer through a network, and wherein the at least one processor creates a plurality of independently separated expert module models by matching the obtained expert model characteristic information to the corresponding expert models, and stores the generated plurality of expert module models in a database; and wherein the at least one processor analyzes input data received from a user terminal to obtain domain information corresponding to the input data. The method comprises: a step in which at least one processor determines, based on the acquired domain information and the databased plurality of specialized model characteristic information, at least one specialized module model having characteristics corresponding to the domain information as a domain-specific specialized model; a step in which at least one processor constructs a domain-specific MoE model by combining the determined at least one domain-specific specialized model and the router; and a step in which at least one processor generates and provides output data for the input data using the constructed domain-specific MoE model.","assignee":"주식회사 Lg 경영개발원","inventors":["최예묵"],"publication_date":"2026-04-09","filing_date":"2026-04-02","priority_date":"2024-05-17","cpc_codes":["G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06N","G06N5/00","G06N5/04","G06N5/043","G","G06","G06N","G06N7/00","G06N7/01"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260048248A/en"},{"publication_number":"KR20260048241A","title":"Method for determining a user's concentration based on an artificial neural network using concentration scores obtained from beta and theta waves measured by a brainwave device as training data and operating server performing the same","abstract":"뇌파 기기를 이용해 측정한 베타파와 세타파 이용해 얻어진 집중력 점수를 학습 데이터로 이용하여 인공신경망 기반의 사용자의 집중력을 판단하는 방법 및 이를 수행하는 운영 서버가 개시된다. 상기 집중력 스코어링 방법을 수행하는 운영 서버는, 적어도 하나의 프로세서(processor); 및 상기 적어도 하나의 프로세서가 적어도 하나의 동작(operation)을 수행하도록 지시하는 명령어들(instructions)을 저장하는 메모리(memory)를 포함하고, 상기 적어도 하나의 동작은, 사용자를 촬영한 사용자 이미지를 획득하는 동작; 상기 사용자 이미지로부터 상기 사용자의 머리에 대한 수평 방향 움직임 각도와 수직 방향 움직임 각도를 결정하고, 상기 수평 방향 움직임 각도와 상기 수직 방향 움직임 각도를 이용하여 머리 움직임 점수를 산출하는 동작; 상기 사용자 이미지로부터 상기 사용자의 시선 방향을 지시하는 시선 벡터를 결정하고, 결정된 상기 시선 벡터를 이용하여 상기 사용자의 시선 움직임 점수를 산출하는 동작; 상기 사용자 이미지로부터 상기 사용자의 눈을 식별하고, 식별된 상기 사용자의 눈을 이용하여 눈감김 점수를 산출하는 동작; 상기 머리 움직임 점수, 상기 시선 움직임 점수, 및 상기 눈감김 점수를 이용하여 입력 데이터를 생성하는 동작; 상기 입력 데이터를 미리 지도학습된 딥러닝 기반의 스코어링 학습 모델에 입력하는 동작; 및 상기 스코어링 학습 모델의 출력으로 집중력 점수를 획득하는 동작;을 포함한다. A method for determining a user's concentration based on an artificial neural network using concentration scores obtained using beta waves and theta waves measured by a brainwave device as training data, and an operating server for performing the same are disclosed. The operating server for performing the concentration scoring method comprises: at least one processor; and a memory storing instructions that instruct the at least one processor to perform at least one operation. The at least one operation comprises: acquiring a user image of a user; determining a horizontal movement angle and a vertical movement angle for the user's head from the user image, and calculating a head movement score using the horizontal movement angle and the vertical movement angle; determining a gaze vector indicating the direction of the user's gaze from the user image, and calculating a gaze movement score using the determined gaze vector; identifying the user's eyes from the user image, and calculating an eye closure score using the identified user's eyes; and generating input data using the head movement score, the gaze movement score, and the eye closure score. The operation of inputting the above input data into a pre-supervised deep learning-based scoring learning model; and the operation of obtaining an attention score as the output of the scoring learning model; are included.","assignee":"(주)대교씨엔에스","inventors":["윤희동","이상진","이세욱","박봉석","문동규"],"publication_date":"2026-04-09","filing_date":"2026-03-31","priority_date":"2023-01-26","cpc_codes":["A","A61","A61B","A61B5/00","A61B5/16","A61B5/168","A","A61","A61B","A61B5/00","A61B5/103","A61B5/11","A61B5/1116","A","A61","A61B","A61B5/00","A61B5/103","A61B5/11","A61B5/1126","A61B5/1128","A","A61","A61B","A61B5/00","A61B5/16","A61B5/163","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/369","A61B5/372","A61B5/374","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/70"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260048241A/en"},{"publication_number":"AU2026202306A1","title":"Data protection via attributes-based aggregation","abstract":"Systems and methods for obfuscating sensitive data by aggregating the data based on data attributes are provided. Each sensitive data record contains at least one sensitive attribute. A data protection system generates a data transformation model based on the sensitive data records and transforms the sensitive data records using the data transformation model. The data protection system further compresses the sensitive database by grouping the sensitive data records into aggregation segments based on the transformed sensitive data records. The data protection system generates aggregated data by calculating statistics for the at least one sensitive attribute contained in the sensitive data records in each of the aggregation segments. The aggregated data can be made accessible by client computing devices that are unauthorized to access the sensitive data records.","assignee":"Equifax Inc","inventors":["Xinyu MIN","Rupesh Ramanlal PATEL"],"publication_date":"2026-04-09","filing_date":"2026-03-25","priority_date":"2019-05-14","cpc_codes":["H","H04","H04L","H04L63/00","H04L63/04","G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6227","G","G06","G06F","G06F21/00","G06F21/60","G06F21/602","G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6245","G","G06","G06N","G06N20/00","H","H04","H04W","H04W12/00","H04W12/02"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202306A1/en"},{"publication_number":"KR20260048223A","title":"System and method for generating an AI persona using control parameters corresponding to MBTI personality type indicators","abstract":"사용자의 성향 또는 설정된 목적에 따라 인공지능의 반응 특성을 제어하는 시스템에 있어서, 외향(E)-내향(I), 감각(S)-직관(N), 사고(T)-감정(F), 판단(J)-인식(P)을 포함하는 8가지 MBTI 성격 측면(Aspect) 중 적어도 하나 이상에 대응하는 제어 파라미터를 저장하는 데이터베이스; 입력된 데이터로부터 사용자의 의도를 분석하는 의도 분석부; 및 상기 분석된 의도에 대하여, 상기 데이터베이스에서 선택된 특정 성격 측면의 파라미터를 적용하여 응답 데이터의 출력 방식(출력량, 논리적 근거 비중, 공감 문구 포함 여부, 구조화 정도 중 적어도 하나 이상)을 결정하는 응답 생성부를 포함하며, 상기 응답 생성부는 상기 8가지 측면의 조합에 의해 정의된 특정 성격 유형에 부합하도록 인공지능의 페르소나를 동적으로 구성하는 것을 특징으로 하는 인공지능 페르소나 생성 시스템이 제공된다. An AI persona generation system is provided, comprising: a database storing control parameters corresponding to at least one of eight MBTI personality aspects including extroversion (E)-introversion (I), sensing (S)-intuition (N), thinking (T)-feeling (F), and judging (J)-perceiving (P); an intention analysis unit that analyzes the user's intention from input data; and a response generation unit that determines the output method of response data (at least one of output volume, logical basis weight, inclusion of empathy phrases, and degree of structuring) by applying parameters of a specific personality aspect selected from the database to the analyzed intention, wherein the response generation unit dynamically configures the AI persona to match a specific personality type defined by a combination of the eight aspects.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-09","filing_date":"2026-03-23","priority_date":"2026-03-23","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/335","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F40/00","G06F40/30","G06F40/35","G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260048223A/en"},{"publication_number":"AU2026202178A1","title":"Digital modeling and tracking of agricultural fields for implementing agricultural field trials","abstract":"A system for implementing a trial in one or more fields is provided. In an embodiment, a server computer receives field data for a plurality of agricultural fields. Based, at least in part, on the field data for the plurality of agricultural fields, the server computer identifies one or more target agricultural fields. The server computer sends, to a field manager computing device associated with the one or more target agricultural fields, a trial participation request. The server receives data indicating acceptance of the trial participation request from the field manager computing device. The server determines one or more locations on the one or more target agricultural fields for implementing a trial and sends data identifying the one or more locations to the field manager computing device. When the server computer receives application data for the one or more target agricultural fields, the server computer determines whether the one or more target agricultural fields are in compliance with the trial. The server computer then receives result data for the trial and, based on the result data, computes a benefit value for the trial.","assignee":"Climate LLC","inventors":["Jason Kendrick BULL","Nicholas Charles Cizek","Brandon RINKENBERGER","Thomas Gene Ruff","Doug SAUDER"],"publication_date":"2026-04-09","filing_date":"2026-03-20","priority_date":"2017-08-21","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/02","A","A01","A01B","A01B79/00","A01B79/005","F","F24","F24F","F24F11/00","F24F11/30","F24F11/49","G","G01","G01C","G01C11/00","G01C11/02","G01C11/025","G","G01","G01N","G01N33/00","G01N33/24","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06393","A","A01","A01B","A01B69/00","G","G06","G06N","G06N3/00","G06N3/12","G06N3/126","Y","Y02","Y02A","Y02A40/00","Y02A40/10"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202178A1/en"},{"publication_number":"AU2026202153A1","title":"Method of selecting a user profile of an application for opening a url","abstract":"The present invention relates to a computer-implemented method of selecting a user profile of an application by a control application implemented on a user device for opening a URL and opening, by the application, the URL using the selected user profile of the application.","assignee":"11point2 Pty Ltd","inventors":["George FRENEY","Ashleigh Greaves","Rajat Kulshrestha","Mark Ogden","Nikhil Singh"],"publication_date":"2026-04-09","filing_date":"2026-03-19","priority_date":"2023-02-10","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G06F16/9535","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/955","G06F16/9566","G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6245","G06F21/6263","G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/445","G06F9/44505","G06F9/4451","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/48","G","G06","G06V","G06V20/00","G06V20/60","G06V20/62","H","H04","H04L","H04L63/00","H04L63/10","H04L63/108","H","H04","H04L","H04L63/00","H04L63/16","H","H04","H04L","H04L67/00","H04L67/01","H04L67/02","H","H04","H04L","H04L67/00","H04L67/2866","H04L67/30","H04L67/306","G","G06","G06F","G06F2221/00","G06F2221/21","G06F2221/2105","G","G06","G06F","G06F2221/00","G06F2221/21","G06F2221/2149","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06V","G06V30/00","G06V30/10","G06V30/19","H","H04","H04L","H04L63/00","H04L63/10","H04L63/102"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202153A1/en"},{"publication_number":"AU2026202097A1","title":"Performing a calibration process in a quantum computing system","abstract":"A B S T R A C T A method comprising: identifying domains of a quantum computing system by operation of a control system, the domains comprising respective domain control subsystems and respective subsets of quantum circuit devices in a quantum processor of the quantum computing system; obtaining a first set of measurements from a first domain of the domains; determining, by operation of the control system, device characteristics of the quantum circuit devices of the first domain based on the first set of measurements; determining to obtain a second set of measurements from the first domain based on the device characteristics; obtaining the second set of measurements from the first domain; determining, by operation of the control system, quantum logic control parameters for the subset of quantum circuit devices of the first domain based on the second set of measurements; and storing the quantum logic control parameters in a database on a special purpose logic circuitry controller of the control system for use in operating the first domain, wherein the special purpose logic circuitry controller has low- latency communication with the quantum processor. 20 26 20 20 97 18 M ar 2 02 6 A B S T R A C T 2 0 2 6 2 0 2 0 9 7 1 8 M a r 2 0 2 6","assignee":"Rigetti and Co LLC","inventors":["Nasser Alidoust","Benjamin Jacob BLOOM","Shane Arthur CALDWELL","Michael James Curtis","Peter Jonathan KARALEKAS","Matthew J. REAGOR","Chad Tyler RIGETTI","Eyob A. SETE","Nikolas Anton TEZAK","William J. ZENG"],"publication_date":"2026-04-09","filing_date":"2026-03-18","priority_date":"2017-03-10","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/70","G","G06","G06N","G06N10/00","G","G06","G06N","G06N10/00","G06N10/40","H","H03","H03K","H03K19/00","H03K19/02","H03K19/195","G","G06","G06N","G06N10/00","G06N10/20"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026202097A1/en"},{"publication_number":"KR20260047556A","title":"System for managing safety linked to cognitive information based image including module","abstract":"작업현장에서 기존의 영상 모니터링에 활용되는 일반 CCTV(Closed Circuit Television) 시스템만을 활용하여 추가 감지센서 및 장비 사용 없이 CCTV의 촬영 영상에 대한 영상인식을 통하여 위험 및 주의 상황을 인지할 수 있는 영상기반 인지정보를 연계한 안전 관리 시스템이 개시된다. 상기 안전 관리 시스템은, 작업장의 영상 정보를 수집하는 영상 수집부, 수집된 상기 영상 정보에 대하여 기계 학습을 통해 영상잡음을 분류하여 상황 인지 정보를 생성하고 상기 영상잡음을 제거하여 잡음 제거 영상 정보를 생성하는 영상 처리부, 및 상기 상황 인지 정보의 실시간 모니터링을 통해 상기 작업장의 환경 농도를 판단하여 상기 작업장의 조명 또는 환기 설비를 자동으로 제어하는 제어 관리부를 포함하는 것을 특징으로 한다. A safety management system linked with image-based recognition information is disclosed, which can recognize dangerous and cautionary situations through image recognition of captured CCTV footage without using additional detection sensors or equipment, by utilizing only a standard CCTV (Closed Circuit Television) system used for existing video monitoring at a work site. The safety management system is characterized by comprising: an image collection unit that collects image information of a workplace; an image processing unit that generates situational recognition information by classifying image noise through machine learning on the collected image information and generates noise-removed image information by removing the image noise; and a control management unit that determines the environmental concentration of the workplace through real-time monitoring of the situational recognition information and automatically controls the lighting or ventilation facilities of the workplace.","assignee":"한국전력공사","inventors":["박준범","박윤식","박예슬","김대현","김영준"],"publication_date":"2026-04-08","filing_date":"2026-04-02","priority_date":"2021-11-25","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06F","G06F18/00","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06T","G06T5/00","G06T5/70","G","G06","G06T","G06T7/00","G06T7/10","G06T7/13"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260047556A/en"},{"publication_number":"EP4721000A2","title":"Multi-model augmented-reality instructions for assembly","abstract":"Augmented-reality (AR) instructions for assembly of a physical object may be based on an assembly sequence divided into multiple assembly phases, using separate computational three-dimensional (3D) registration models (e.g., object recognition models) for the different phases. A user may be guided step by step through the assembly sequence, using a step counter to keep track of where in the assembly sequence the user is. In each step, an AR image may be created by rendering, overlaid onto the camera image, a virtual model of a part of the physical object to be added during the step, spatially registered with the physical object in its partially assembled state.","assignee":"Texas A&M University System","inventors":["Wei Yan","Seda TUZUN CANADINC"],"publication_date":"2026-04-08","filing_date":"2024-06-02","priority_date":"2023-06-02","cpc_codes":["G","G09","G09B","G09B5/00","G09B5/02","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G09","G09B","G09B19/00","G09B19/003","G","G09","G09B","G09B19/00","G09B19/0069","G","G09","G09B","G09B5/00","G09B5/08","G09B5/12","G09B5/125"],"country":"EP","kind":"application","source_url":"https://patents.google.com/patent/EP4721000A2/en"},{"publication_number":"KR20260046355A","title":"System and method for providing 3rd profit generation service for influencers in social media platform","abstract":"본 발명은 소셜 미디어 플랫폼에서 인플루언서의 제3 수익을 창출하는 서비스 제공 시스템 및 방법에 관한 것이다. 본 시스템은 미리 결정된 사용자 이벤트 버튼의 구동에 따라, 상기 소셜 미디어 플랫폼에서 제공되는 상기 인플루언서의 컨텐츠 정보를 외부로 제공하는 사용자 단말; 및 상기 사용자 단말로부터 상기 컨텐츠 정보를 제공받아 상기 인플루언서의 컨텐츠로부터 미리 결정된 유형의 디텍팅 정보를 추출하고, 상기 디텍팅 정보를 기반으로 상기 제3 수익을 창출하기 위한 광고 유통 플랫폼을 생성하여 상기 사용자 단말에 제공하는 서버를 포함하고, 상기 광고 유통 플랫폼이 상기 사용자 단말에서 실행되어 거래에 의한 수익이 발생하는 경우, 상기 인플루언서에게 상기 수익의 적어도 일부가 제공되는 것을 특징으로 한다. The present invention relates to a service provision system and method for generating third-party revenue for an influencer on a social media platform. The present system comprises: a user terminal that provides content information of the influencer provided on the social media platform to the outside upon activation of a predetermined user event button; and a server that receives the content information from the user terminal, extracts detection information of a predetermined type from the influencer's content, and creates an advertising distribution platform for generating the third revenue based on the detection information and provides it to the user terminal, wherein when the advertising distribution platform is executed on the user terminal and revenue is generated through a transaction, at least a portion of the revenue is provided to the influencer.","assignee":"(주)캔버시; 모영일; 이숙경","inventors":["모영일","이숙경"],"publication_date":"2026-04-07","filing_date":"2026-03-30","priority_date":"2019-09-06","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0251","G06Q30/0269","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/04","G06N5/045","G","G06","G06Q","G06Q10/00","G06Q10/40","G","G06","G06Q","G06Q10/00","G06Q10/40","G06Q10/46","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0207","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0251","G06Q30/0267","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0273","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0276","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0281","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0631","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/08","G","G06","G06Q","G06Q50/00","G06Q50/10"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260046355A/en"},{"publication_number":"KR20260046347A","title":"Method for predicting performance on image to be evaluated using neural network model","abstract":"본 개시의 일 실시예에 따라, 신경망 모델을 활용하여 평가 대상 이미지에 대한 성과를 예측하기 위한 방법이 개시된다. 상기 방법은, 평가 대상 이미지와 관련된 복수의 정량화된 지표들 중 하나 이상의 지표를 결정하는 단계; 상기 평가 대상 이미지의 복수의 이미지-관련 특징들 중 하나 이상의 이미지-관련 특징을 결정하는 단계; 상기 결정된 하나 이상의 지표 및 상기 결정된 하나 이상의 이미지-관련 특징에 기초하여, 상기 평가 대상 이미지에 대한 제 1 잠재 벡터(latent vector)를 추출하는 단계; 및 상기 추출된 제 1 잠재 벡터에 기초하여, 상기 평가 대상 이미지에 대한 성과를 예측하는 단계를 포함할 수 있다. According to one embodiment of the present disclosure, a method for predicting performance for an image to be evaluated using a neural network model is disclosed. The method may include: determining one or more indicators among a plurality of quantified indicators related to an image to be evaluated; determining one or more image-related features among a plurality of image-related features of an image to be evaluated; extracting a first latent vector for the image to be evaluated based on the one or more determined indicators and the one or more determined image-related features; and predicting performance for the image to be evaluated based on the extracted first latent vector.","assignee":"주식회사 테이아","inventors":["김서진","이보형"],"publication_date":"2026-04-07","filing_date":"2026-03-26","priority_date":"2022-08-05","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0242","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/951","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06393","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0276","G","G06","G06T","G06T7/00","G06T7/97","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/761"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260046347A/en"},{"publication_number":"KR20260046343A","title":"Anomaly detection method and appratus using neural network","abstract":"뉴럴 네트워크를 이용하여 이상치를 탐지하는 전자 장치가 개시된다. 일 실시예에 따른 전자 장치는 메모리, 송수신기, 및 프로세서를 포함하고, 상기 프로세서는, 입력 이미지에 기초하여 산출된 심층 특징(feature) 및 상기 심층 특징의 평균에 기초하여 상기 입력 이미지에 대응하는 이상치를 탐지하도록 미리 학습된 뉴럴 네트워크 모델에 기초하여, 상기 입력 이미지의 이상치를 탐지할 수 있다. An electronic device for detecting outliers using a neural network is disclosed. An electronic device according to one embodiment includes a memory, a transceiver, and a processor, wherein the processor can detect outliers in an input image based on a neural network model that is pre-trained to detect outliers corresponding to the input image based on deep features calculated based on the input image and the average of said deep features.","assignee":"주식회사 엘로이랩","inventors":["유광선","윤혁"],"publication_date":"2026-04-07","filing_date":"2026-03-25","priority_date":"2022-05-23","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2433","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T7/00","G06T7/0002","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260046343A/en"},{"publication_number":"KR20260047202A","title":"Hybrid operating system of local and cloud intelligent engines based on user prompt analysis and the operating method thereof","abstract":"본 발명은 개인용 컴퓨팅 장치 내에서 AI 에이전트를 운용함에 있어, 사용자의 프롬프트 명령을 다차원적으로 분석하여 보안성이 중요한 작업은 장치 내 로컬 AI 엔진인 제1 지능형 엔진이 처리하고, 높은 지능 및 추론 능력이 필요한 작업은 외부 클라우드 AI 엔진인 제2 지능형 엔진이 처리하도록동적으로 제어하는 지능형 실행 엔진을 구비한 하이브리드 운용 시스템 및 그 동작 방법에 관한 것이다. 상기 지능형 실행 엔진은 프롬프트 수신부, 프롬프트 분석기, 엔진 선택 결정부, 보안 판단 모듈, 복잡도 판단 모듈, 비용 최적화 모듈, 학습 DB, 검증부 및 네트워크 모니터링부를 포함하며, 클라우드 엔진이 수립한 전략을 로컬 엔진이 검증한 후 실행하는 이중 안전 구조를 통해 보안성과 지능성을 동시에 달성한다. 이를 통해 사용자는 API 시크릿 키 등 민감 데이터의 외부 유출 없이 고성능 클라우드 AI의 혜택을 누리며 안전하게 시스템 제어 및 자동화 작업을 수행할 수 있다. The present invention relates to a hybrid operating system and a method of operation thereof, comprising an intelligent execution engine that dynamically controls the operation of an AI agent within a personal computing device by multidimensionally analyzing user prompt commands, wherein tasks critical to security are processed by a first intelligent engine, which is a local AI engine within the device, and tasks requiring high intelligence and reasoning capabilities are processed by a second intelligent engine, which is an external cloud AI engine. The intelligent execution engine includes a prompt receiving unit, a prompt analyzer, an engine selection decision unit, a security judgment module, a complexity judgment module, a cost optimization module, a training DB, a verification unit, and a network monitoring unit, and achieves both security and intelligence simultaneously through a dual-safety structure in which the local engine verifies and executes a strategy established by the cloud engine. Through this, users can safely perform system control and automation tasks while enjoying the benefits of high-performance cloud AI without the external leakage of sensitive data, such as API secret keys.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-07","filing_date":"2026-03-20","priority_date":"2026-03-20","cpc_codes":["H","H04","H04L","H04L67/00","H04L67/50","H04L67/60","H04L67/63","G","G06","G06N","G06N3/00","G06N3/02","G06N3/10"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260047202A/en"},{"publication_number":"KR20260047204A","title":"Smart Glasses System for Selective Hybrid AI Computing Based on App Attributes and Method for Controlling the Same","abstract":"본 발명은 스마트 안경 장치(100)에서 실행되는 어플리케이션의 속성, 즉 요구 지연(Latency), 요구 연산량(FLOPS), 데이터 보안 등급, 및 네트워크 접속 상태를 앱 속성 분석 모듈(131)이 실시간으로 판별하고, 자원 선택 모듈(132)이 판별된 속성에 기초하여 독립 전원부(260) 및 NPU(210)를 갖춘 로컬 지능형 AI 컴퓨팅 장치(200)와 클라우드 기반 거대 지능형 AI 서버(300) 중 최적의 연산 주체를 동적으로 선택하는 하이브리드 AI 컴퓨팅 선택형 스마트 안경 시스템 및 그 제어 방법을 개시한다. 특히 보안 관리 모듈(133)은 개인 식별 정보를 포함하는 민감 데이터의 클라우드 전송을 구조적으로 차단하여 데이터 프라이버시를 보호하고, 자원 모니터링 모듈(134)은 로컬 지능형 AI 컴퓨팅 장치(200)의 부하를 실시간 감시하여 초과 시 연산을 클라우드 기반 거대 지능형 AI 서버(300)로 자동 분산함으로써 서비스 연속성을 보장한다. 본 발명에 따르면 착용형 스마트 안경의 배터리 소모를 최소화하면서 클라우드에 준하는 고성능 AI 연산을 초저지연 로컬 환경에서 제공할 수 있다. The present invention discloses a hybrid AI computing selectable smart glasses system and a control method thereof, wherein an app attribute analysis module (131) determines in real time the attributes of an application running on a smart glasses device (100), namely latency, FLOPS, data security level, and network connection status, and a resource selection module (132) dynamically selects the optimal computing entity between a local intelligent AI computing device (200) equipped with an independent power supply (260) and an NPU (210) and a cloud-based large intelligent AI server (300) based on the determined attributes. In particular, a security management module (133) protects data privacy by structurally blocking the transmission of sensitive data containing personal identification information to the cloud, and a resource monitoring module (134) ensures service continuity by monitoring the load of the local intelligent AI computing device (200) in real time and automatically distributing the computation to the cloud-based large intelligent AI server (300) when the load is exceeded. According to the present invention, high-performance AI computation comparable to that of the cloud can be provided in an ultra-low latency local environment while minimizing battery consumption of the wearable smart glasses.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-07","filing_date":"2026-03-20","priority_date":"2026-03-20","cpc_codes":["G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481","G06F3/04815","G","G02","G02B","G02B27/00","G02B27/01","G02B27/017","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06T","G06T19/00","G06T19/006","H","H04","H04W","H04W4/00","H04W4/80","G","G02","G02B","G02B27/00","G02B27/01","G02B27/017","G02B2027/0178"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260047204A/en"},{"publication_number":"KR20260047199A","title":"Artificial Intelligence-Based Religious Content Provision System Including a Multi-Religion Switching Algorithm, Noise Filtering, and User-Driven Spiritual Curation","abstract":"본 발명은 인공지능 기반 종교 콘텐츠 제공 시스템에 관한 것으로서, 사용자로부터 입력된 질문, 고민 또는 종교 활동 관련 데이터를 분석하여 복수 종교의 텍스트 데이터베이스 중 사용자 상황에 적합한 종교 콘텐츠를 선택하는 다종교 스위칭 알고리즘을 포함하는 종교 콘텐츠 제공 시스템에 관한 것이다. 본 발명의 시스템은 사용자 입력 데이터를 분석하는 의미 분석 엔진, 복수 종교의 텍스트 데이터베이스를 참조하여 종교 콘텐츠 후보를 추출하는 데이터 처리 모듈, 사용자 입력 데이터와 종교 텍스트 간 의미 유사도를 계산하여 종교 콘텐츠를 선택하는 다종교 스위칭 엔진을 포함할 수 있다. 또한 본 발명은 선택된 종교 콘텐츠에 포함된 상업적 메시지, 정치적 메시지, 폭력적 표현, 비방 또는 모함 표현 및 허위 정보를 제거하는 노이즈 필터링 기능과 사용자가 종교 유형 또는 콘텐츠 유형을 직접 선택할 수 있도록 하는 사용자 주도형 영적 큐레이션 기능을 포함할 수 있다. 이를 통해 사용자 상황에 적합한 종교 콘텐츠를 제공하고 복수 종교 또는 성현 및 철학자의 지혜 텍스트를 기반으로 한 맞춤형 종교 콘텐츠 제공이 가능하도록 하는 효과가 있다. The present invention relates to an artificial intelligence-based religious content provision system, comprising a multi-religion switching algorithm that analyzes data related to questions, concerns, or religious activities input by a user and selects religious content suitable for the user's situation from a text database of multiple religions. The system of the present invention may include a semantic analysis engine that analyzes user input data, a data processing module that extracts religious content candidates by referring to a text database of multiple religions, and a multi-religion switching engine that selects religious content by calculating semantic similarity between user input data and religious text. In addition, the present invention may include a noise filtering function that removes commercial messages, political messages, violent expressions, slanderous or defamatory expressions, and false information included in selected religious content, and a user-driven spiritual curation function that allows the user to directly select a type of religion or a type of content. This has the effect of providing religious content suitable for the user's situation and enabling the provision of customized religious content based on texts of multiple religions or the wisdom of sages and philosophers.","assignee":"최규옥","inventors":["최규옥"],"publication_date":"2026-04-07","filing_date":"2026-03-19","priority_date":"2026-03-19","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06F","G06F16/00","G06F16/20","G06F16/22","G06F16/2228","G06F16/2237","G","G06","G06F","G06F16/00","G06F16/60","G06F16/64","G","G06","G06F","G06F16/00","G06F16/90","G06F16/906","G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260047199A/en"},{"publication_number":"KR20260046064A","title":"Lifetime estimation system, lifetime estimation method and computer-readable storage media","abstract":"본 발명에 관한 수명 추정 시스템은, 건설 기계의 경사 각도를 측정하는 경사 측정부와, 상기 건설 기계의 주행 조작에 따라 변화하는 주행 제어량을 측정하는 제어량 측정부와, 상기 경사 측정부에 의해 측정된 상기 경사 각도와 상기 제어량 측정부에 의해 검출된 상기 주행 제어량을 관련지어 이력 정보로서 기억하는 기억부와, 상기 기억부에 기억되어 있는 상기 이력 정보에 기초하여, 상기 건설 기계의 주행 부품의 수명을 추정하는 추정부와, 상기 추정부에 의한 추정값을 출력하는 출력부를 구비한다. A life estimation system according to the present invention comprises: an inclination measuring unit for measuring the inclination angle of a construction machine; a control amount measuring unit for measuring a driving control amount that changes according to the driving operation of the construction machine; a memory unit for storing historical information by associating the inclination angle measured by the inclination measuring unit with the driving control amount detected by the control amount measuring unit; an estimation unit for estimating the life of a driving component of the construction machine based on the historical information stored in the memory unit; and an output unit for outputting an estimated value by the estimation unit.","assignee":"나부테스코 가부시키가이샤","inventors":["겐타 하세가와"],"publication_date":"2026-04-06","filing_date":"2026-03-25","priority_date":"2021-12-03","cpc_codes":["G","G01","G01M","G01M17/00","G01M17/007","E","E02","E02F","E02F9/00","E02F9/26","E02F9/264","G","G01","G01D","G01D21/00","G01D21/02","G","G01","G01L","G01L5/00","G01L5/0061","G01L5/0071","G","G01","G01M","G01M5/00","G01M5/0033","G","G01","G01N","G01N33/00","G01N33/0078","G01N33/0083","G","G01","G01R","G01R19/00","G01R19/0092","G","G05","G05B","G05B13/00","G05B13/02","G05B13/0205","G05B13/026","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","B","B60","B60Y","B60Y2200/00","B60Y2200/40","B60Y2200/41"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260046064A/en"},{"publication_number":"KR20260046062A","title":"Information sharing system for co-parenting","abstract":"본 발명은 공동 양육을 위한 정보 공유 시스템에 관한 것이다. 본 발명에 따르면, 정보 공유 시스템은 네트워크 접속된 사용자 단말에 실행된 앱을 통하여 식사, 수면, 배변 및 질병 진단 여부 중에서 적어도 생활 데이터를 입력받는 입력부; 생활 데이터와 표준 데이터 수집부에 저장된 데이터를 비교하여 해당되는 아이의 식사량, 수면량 및 배변량이 적정한지 여부를 판단하고, 판단 결과에 따른 적정한 식사량, 수면량 및 배변량을 산출하는 산출부; 및 산출된 식사량, 수면량 및 배변량에 따른 정보를 제공하고, 양육자로부터 아이의 행동 발달 또는 훈육 및 학습 관련한 데이터를 요청 받을 경우, 동일한 연령대의 표준데이터를 기반으로 정보를 제공하는 솔루션 제공부;를 포함한다. 본 발명에 따르면, 아이를 육아하는 모든 양육자들에게 데이터 접근 권한을부여하여 아이에 대한 정보를 공유할 수 있으므로 공동 육아 효과를 증대시킬 수 있다. 특히, 주 양육자가 아니더라도 아이를 돌보는 주체가 사용자 단말을 통해서 생활 데이터를 직접 입력하게 하여 주 양육자의 부감을 감소시키고, 아이의 케어에 있어서 일관성과 정확성을 증대시켜 안정감을 도모할 수 있다. The present invention relates to an information sharing system for co-parenting. According to the present invention, the information sharing system comprises: an input unit that receives at least lifestyle data among meals, sleep, bowel movements, and disease diagnosis status through an app executed on a network-connected user terminal; a calculation unit that compares lifestyle data with data stored in a standard data collection unit to determine whether the corresponding child's meal amount, sleep amount, and bowel movement amount are appropriate, and calculates the appropriate meal amount, sleep amount, and bowel movement amount based on the determination result; and a solution providing unit that provides information based on the calculated meal amount, sleep amount, and bowel movement amount, and, when a caregiver requests data related to the child's behavioral development or discipline and learning, provides information based on standard data of the same age group. According to the present invention, by granting data access rights to all caregivers raising a child, information about the child can be shared, thereby enhancing the effects of co-parenting. In particular, by allowing anyone caring for the child—even those who are not the primary caregiver—to directly input daily life data through a user terminal, the burden on the primary caregiver can be reduced, and a sense of stability can be promoted by increasing consistency and accuracy in child care.","assignee":"조정원","inventors":["조정원"],"publication_date":"2026-04-06","filing_date":"2026-03-20","priority_date":"2023-12-11","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/22","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/40","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G16","G16H","G16H50/00","G16H50/70"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260046062A/en"},{"publication_number":"KR20260046059A","title":"Personal computing device comprising an on-device execution agent interfaced with a generative AI server","abstract":"본 발명은 외부의 생성형 AI 서버와 연동하여 로컬 작업을 수행하는 개인용 컴퓨팅 장치에 관한 것이다. 본 발명의 개인용 컴퓨팅 장치는 자연어 프롬프트를 입력받는 입력부, 프롬프트를 AI 서버로 전송하고 자유형식 자연어 답변 데이터를 수신하는 데이터 송수신부, 및 개인용 컴퓨팅 장치 내부에 온-디바이스 방식으로 탑재된 실행 에이전트를 포함한다. 실행 에이전트는 자연어 답변 데이터를 의미 분석하여 제어 대상과 파라미터를 식별하는 파싱 모듈, 설치된 운영체제의 종류를 판별하여 OS 네이티브 시스템 호출 코드 또는 제어 스크립트로 변환하는 OS 적응형 변환 모듈, 및 변환된 코드를 GUI 조작 없이 OS 커널 레벨에서 직접 실행하는 네이티브 실행 모듈로 구성된다. 본 발명에 의하면 GUI 모사 방식 대비 실행 지연이 현저히 단축되고, 오프라인 환경에서도 캐시를 통한 동작이 가능하며, 보안 검증 모듈을 통하여 악의적 시스템 호출을 사전에 차단할 수 있다. The present invention relates to a personal computing device that performs local tasks by linking with an external generative AI server. The personal computing device of the present invention includes an input unit that receives a natural language prompt, a data transmission and reception unit that transmits the prompt to an AI server and receives free-form natural language response data, and an execution agent mounted on-device within the personal computing device. The execution agent is composed of a parsing module that identifies control targets and parameters by semantically analyzing the natural language response data, an OS adaptive conversion module that determines the type of installed operating system and converts it into OS native system call code or control script, and a native execution module that directly executes the converted code at the OS kernel level without GUI manipulation. According to the present invention, execution delay is significantly reduced compared to GUI simulation methods, operation via caching is possible even in offline environments, and malicious system calls can be blocked in advance through a security verification module.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-06","filing_date":"2026-03-18","priority_date":"2026-03-18","cpc_codes":["G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/445","G06F9/44568","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/4401","G06F9/4406","G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/445","G06F9/44505","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260046059A/en"},{"publication_number":"KR20260046054A","title":"AI Agent System with Data Refining Function for Vision-based User Behavior Imitation Learning and Method Thereof","abstract":"본 발명은 비전 기반 사용자 행동 모방 학습을 위한 데이터 정제 기능을 구비한 AI 에이전트 시스템 및 그 방법에 관한 것이다. 본 발명의 AI 에이전트 시스템(100)은 사용자의 컴퓨터 조작 화면 이미지를 실시간 캡처하는 비전 수집부(110), 입력 이벤트를 수집하는 입력 이벤트 감지부(120), 및 제어부(130)를 포함한다. 제어부(130)는 목표 상태 분석기(131), 노이즈 분석 모듈(132), 행동 시퀀스 생성기(133), 및 경로 재구성부(134)를포함하며, 노이즈 분석 모듈(132)은 세만틱 필터(132a)에 의한 1차 노이즈 분류, 비주얼 델타 분석기(132b)에 의한 2차 노이즈 분류, 및 그래프 탐색 엔진(132c)에 의한 3차 노이즈 분류를 순차적으로 수행하여 목표 작업과 무관한 행동을 자동 제거한다. 경로 재구성부(134)는 노이즈 제거 후 불연속 구간을 스티칭 처리하여 최적화된 자동화 워크플로우를 완성한다. 본 발명에 의하면 AI 에이전트의 학습 효율과 자동화 재현 안정성이 향상되고, 사용자가 자연스럽게 시연하는 것만으로도 고품질의 자동화 학습 데이터가 자동 생성된다. The present invention relates to an AI agent system and a method equipped with a data refinement function for vision-based user behavior imitation learning. The AI agent system (100) of the present invention includes a vision collection unit (110) that captures a user's computer operation screen image in real time, an input event detection unit (120) that collects input events, and a control unit (130). The control unit (130) includes a target state analyzer (131), a noise analysis module (132), a behavior sequence generator (133), and a path reconstruction unit (134). The noise analysis module (132) sequentially performs first-order noise classification by a semantic filter (132a), second-order noise classification by a visual delta analyzer (132b), and third-order noise classification by a graph search engine (132c) to automatically remove behaviors unrelated to the target task. The path reconstruction unit (134) completes an optimized automated workflow by stitching discontinuous sections after noise removal. According to the present invention, the learning efficiency and automated reproduction stability of an AI agent are improved, and high-quality automated learning data is automatically generated simply by the user naturally demonstrating it.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-06","filing_date":"2026-03-18","priority_date":"2026-03-18","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06V","G06V40/00","G06V40/20"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260046054A/en"},{"publication_number":"KR20260045768A","title":"Decoding method for multi-view image based on super-resolution and decoder","abstract":"초해상화를 이용한 다시점 비디오의 복호 방법은 디코더가 인코더에서 다운 샘플링되어 부호화된 타깃 영상의 비트 스트림을 입력받는 단계, 상기 디코더가 저해상도인 상기 타깃 영상을 복호하는 단계 및 상기 디코더가 상기 타깃 영상 및 상기 타깃 영상에 대한 참조 영상을 사전에 학습된 신경망 모델에 입력하여 상기 타깃 영상에 대한 초해상화를 수행하는 단계를 포함한다. A method for decoding multi-view video using super-resolution includes the steps of: a decoder receiving a bitstream of a target image that has been downsampled and encoded by an encoder; the decoder decoding the target image which is of low resolution; and the decoder inputting the target image and a reference image for the target image into a pre-trained neural network model to perform super-resolution on the target image.","assignee":"한국전자기술연구원","inventors":["강제원","김용환"],"publication_date":"2026-04-03","filing_date":"2026-03-23","priority_date":"2021-10-13","cpc_codes":["H","H04","H04N","H04N19/00","H04N19/50","H04N19/597","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T3/00","G06T3/40","G06T3/4046","G","G06","G06T","G06T3/00","G06T3/40","G06T3/4053","G","G06","G06T","G06T9/00","G06T9/002","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/103","H04N19/105","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/124","H","H04","H04N","H04N19/00","H04N19/42","H04N19/423"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260045768A/en"},{"publication_number":"KR20260045759A","title":"Method and AX System for AI Structuring of Unstructured and Semi-structured Data Based on 5W1H Framework","abstract":"본 발명은 5W1H 기반 비정형·반정형 데이터 AI구조화 방법 및 AX시스템에 관한 것이다.본 발명은 비정형 데이터(TXT·DOCX·PDF·이미지) 및반정형 데이터(JSON·XML·로그)를 단일 처리 파이프라인으로 통합 입력받아, 파일 유형 자동 감지 모듈(310)이 최적 처리 경로로 라우팅하고, 텍스트 추출엔진(320)이 유형별 텍스트를 추출하며, 전처리 모듈(330)이 노이즈를 제거한다.이후 5W1H 기반 AI 구조화 엔진(340)이 저널리즘 및언어학 분야에서 100년 이상 검증된 5W1H 이론을적용하여 주체(CH1·341)·내용(CH2·342)·시간(CH3·343)·공간(CH4·344)·목적(CH5·345)·방법(CH6·346)의6가지 의미 채널을 자동 추출하고 표준 JSON 형식으로변환한다. 상기 핵심 알고리즘은 SaaS 서버 내부에캡슐화되어 외부에 노출되지 않는다.품질 평가 모듈(400)은 완전성(410)·정확성(420)·일관성(430)을 자동 평가하여 A·B·C 3단계 품질 등급(440)을 산정하고, AX시스템 구축 모듈(500)은RAG(520) 및 FAISS 벡터 검색 엔진(530)을 활용하여AX(AI Transformation) 시스템을 구축한다.본 발명에 따르면 데이터 처리 시간 약 80% 단축,5W1H 이론 기반 표준화된 의미 구조 추출, 핵심알고리즘 보호, A·B·C 자동 품질 등급 체계 및AX시스템 구축 자동화의 효과를 얻을 수 있다. The present invention relates to a 5W1H-based AI structuring method for unstructured and semi-structured data and an AX system. The present invention receives unstructured data (TXT, DOCX, PDF, images) and semi-structured data (JSON, XML, logs) as integrated inputs in a single processing pipeline, an automatic file type detection module (310) routes to an optimal processing path, a text extraction engine (320) extracts text by type, and a preprocessing module (330) removes noise. Subsequently, a 5W1H-based AI structuring engine (340) applies the 5W1H theory, which has been verified for over 100 years in the fields of journalism and linguistics, to automatically extract six semantic channels of subject (CH1·341), content (CH2·342), time (CH3·343), space (CH4·344), purpose (CH5·345), and method (CH6·346), and converts them into a standard JSON format. The above core algorithm is encapsulated inside the SaaS server and is not exposed externally. The quality evaluation module (400) automatically evaluates completeness (410), accuracy (420), and consistency (430) to calculate a three-level quality grade (440) of A, B, and C, and the AX system construction module (500) utilizes RAG (520) and the FAISS vector search engine (530) to construct an AX (AI Transformation) system. According to the present invention, the effects of reducing data processing time by approximately 80%, extracting standardized semantic structures based on 5W1H theory, protecting core algorithms, an automatic A, B, and C quality grade system, and automating the construction of an AX system can be obtained.","assignee":"배종옥","inventors":["배종옥"],"publication_date":"2026-04-03","filing_date":"2026-03-17","priority_date":"2026-03-17","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/25","G06F16/254","G","G06","G06F","G06F16/00","G06F16/20","G06F16/21","G06F16/211","G","G06","G06F","G06F16/00","G06F16/20","G06F16/22","G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G","G06","G06N","G06N20/00","G","G06","G06V","G06V30/00","G06V30/10","H","H04","H04L","H04L65/00","H04L65/10","H04L65/102","H","H04","H04L","H04L67/00","H04L67/50"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260045759A/en"},{"publication_number":"KR20260045753A","title":"AI-based Integrated Diagnostic Device for Power Facility Installation","abstract":"본 발명은 전력 설비의 운전 중 발생하는 부분방전, 아크, 절연 열화, 침수, 물리적 파손 및 잔여 사용 수명 등을 AI 기반 분석을 통해 통합하여 평가 및 예측할 수 있는 AI를 이용한 전력 설비 설치용 통합 진단장치에 관한 것이다. 상기의 과제를 해결하기 위한 본 발명에 따른 AI를 이용한 전력 설비 설치용 통합 진단장치는, 전류 센서; 광 센서; 전자파 센서; 온도 센서; 부분방전 센서; 진동 센서; 수위 센서; 데이터 수집부; AI연산부; 및 제어부를 포함하고, 상기 AI연산부는 전류 신호에 대해 고속 푸리에 변환(FFT)을 수행하여 고주파 에너지 성분을 계산하고, 광신호의 기울기 및 전자파 신호의 에너지를 산출하며, 수학식 1을 통해 아크 판정 지수를 계산하고, 계산된 아크 판정 지수와 온도 및 아크 발생 횟수를 이용하여 수학식 2를 통해 열화 지수를 계산하며, 시간에 따른 열화 누적 효과를 반영하기 위하여 수학식 3을 통해 열화 누적 지수를 계산하고, 수학식 4를 통해 부분방전 지수를 계산하며, 수학식 5를 통해 진동 영향 지수를 계산하고, 수학식 6을 통해 침수 영향 지수를 계산하며, 계산된 상기 열화 누적 지수와 아크 발생 빈도, 부분방전 지수, 진동 영향 지수, 침수 영향 지수 및 온도 상승률을 입력 변수로 하는 기계학습 기반 수명 예측 모델을 이용하여 잔여 사용 수명을 계산하는 것을 특징으로 한다. The present invention relates to an integrated diagnostic device for power equipment installation using AI, which can integrate and evaluate and predict partial discharge, arc, insulation degradation, water inundation, physical damage, and remaining service life occurring during the operation of power equipment through AI-based analysis. An integrated diagnostic device for power facility installation using AI according to the present invention for solving the above problems comprises: a current sensor; a light sensor; an electromagnetic wave sensor; a temperature sensor; a partial discharge sensor; a vibration sensor; a water level sensor; a data collection unit; and an AI processing unit. The AI computing unit includes a control unit, wherein the AI computing unit performs a Fast Fourier Transform (FFT) on a current signal to calculate a high-frequency energy component, calculates the slope of the optical signal and the energy of the electromagnetic signal, calculates an arc judgment index through Equation 1, calculates a degradation index through Equation 2 using the calculated arc judgment index, temperature, and the number of arc occurrences, calculates a degradation accumulation index through Equation 3 to reflect the degradation accumulation effect over time, calculates a partial discharge index through Equation 4, calculates a vibration influence index through Equation 5, and calculates a water inundation influence index through Equation 6, and calculates the remaining service life using a machine learning-based life prediction model that takes the calculated degradation accumulation index, arc occurrence frequency, partial discharge index, vibration influence index, water inundation influence index, and temperature rise rate as input variables.","assignee":"서규선","inventors":["서규선"],"publication_date":"2026-04-03","filing_date":"2026-03-16","priority_date":"2026-03-16","cpc_codes":["G","G01","G01R","G01R31/00","G01R31/40","G","G01","G01D","G01D21/00","G01D21/02","G","G01","G01F","G01F23/00","G","G01","G01H","G01H11/00","G","G01","G01J","G01J1/00","G01J1/42","G","G01","G01K","G01K1/00","G01K1/02","G","G01","G01R","G01R29/00","G01R29/08","G01R29/0864","G01R29/0892","G","G01","G01R","G01R31/00","G01R31/12","G01R31/1218","G","G06","G06N","G06N20/00","G","G08","G08B","G08B21/00","G08B21/18","G08B21/182","H","H02","H02B","H02B13/00","H02B13/02","H02B13/025"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260045753A/en"},{"publication_number":"KR20260044874A","title":"Switchgear(High-voltage Panel, Low-voltage Panel, Motor Control Panel, Distribution Panel) Equipped with AI-based Partial Discharge, Arc, and Insulation Degradation Diagnosis and Remaining Life Prediction Device","abstract":"본 발명은 부분방전 신호, 전류 신호, 광신호 및 전자파 신호를 종합적으로 분석하여 전력 설비에서 발생하는 부분방전 및 아크를 정확하게 판정하고, AI 알고리즘을 통해 설비의 잔여 사용 수명을 예측할 수 있는 AI를 이용한 부분방전, 아크 및 절연 열화 진단 및 잔여 사용 수명 예측장치가 구비된 수배전반에 관한 것이다. 상기의 과제를 해결하기 위한 본 발명에 따른 AI를 이용한 부분방전, 아크 및 절연 열화 진단 및 잔여 사용 수명 예측장치가 구비된 수배전반은, AI진단장치가 전류 센서; 광 센서; 전자파 센서; 온도 센서; 부분방전 센서; 데이터 수집부; AI연산부; 및 제어부를 포함하고, 상기 AI연산부는 전류 신호에 대해 고속 푸리에 변환(FFT)을 수행하여 고주파 에너지 성분을 계산하고, 광신호의 기울기 및 전자파 신호의 에너지)를 산출하며, 수학식 1을 통해 아크 판정 지수를 계산하고, 계산된 아크 판정 지수와 온도 및 아크 발생 횟수를 이용하여 수학식 2를 통해 열화 지수를 계산하며, 시간에 따른 열화 누적 효과를 반영하기 위하여 수학식 3을 통해 열화 누적 지수를 계산하고, 수학식 4를 통해 부분방전 지수를 계산하며, 계산된 상기 열화 누적 지수와 아크 발생 빈도, 부분방전 지수 및 온도 상승률을 입력 변수로 하는 기계학습 기반 수명 예측 모델을 이용하여 잔여 사용 수명을 계산하는 것을 특징으로 한다. The present invention relates to a switchgear equipped with an AI-based partial discharge, arc, and insulation degradation diagnosis and remaining service life prediction device that comprehensively analyzes partial discharge signals, current signals, optical signals, and electromagnetic signals to accurately determine partial discharge and arc occurring in power equipment, and predicts the remaining service life of the equipment through an AI algorithm. A switchgear equipped with a partial discharge, arc, and insulation degradation diagnosis and remaining service life prediction device using AI according to the present invention for solving the above problems is characterized in that the AI diagnostic device includes a current sensor; an optical sensor; an electromagnetic wave sensor; a temperature sensor; a partial discharge sensor; a data collection unit; an AI computation unit; and a control unit, wherein the AI computation unit performs a Fast Fourier Transform (FFT) on a current signal to calculate a high-frequency energy component and calculates the slope of the optical signal and the energy of the electromagnetic wave signal, calculates an arc judgment index through Equation 1, calculates a degradation index through Equation 2 using the calculated arc judgment index, temperature, and the number of arc occurrences, calculates a degradation accumulation index through Equation 3 to reflect the cumulative degradation effect over time, calculates a partial discharge index through Equation 4, and calculates the remaining service life using a machine learning-based life prediction model that takes the calculated degradation accumulation index, arc occurrence frequency, partial discharge index, and temperature rise rate as input variables.","assignee":"서규선","inventors":["서규선"],"publication_date":"2026-04-02","filing_date":"2026-03-16","priority_date":"2026-03-16","cpc_codes":["G","G01","G01R","G01R31/00","G01R31/40","G","G01","G01D","G01D21/00","G01D21/02","G","G01","G01J","G01J1/00","G01J1/42","G","G01","G01K","G01K1/00","G01K1/02","G","G01","G01R","G01R29/00","G01R29/08","G01R29/0864","G01R29/0892","G","G01","G01R","G01R31/00","G01R31/12","G01R31/1218","G","G06","G06N","G06N20/00","G","G08","G08B","G08B21/00","G08B21/18","G08B21/182","G","G08","G08B","G08B21/00","G08B21/18","G08B21/185","H","H02","H02B","H02B13/00","H02B13/02","H02B13/025"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260044874A/en"},{"publication_number":"AU2026201971A1","title":"System and method for decision support","abstract":"Systems and methods are provided to provide guidance to a user regarding management of a physiologic condition such as diabetes. The determination may be based upon a patient glucose concentration level. The glucose concentration level may be provided to a stored model to determine a state. The guidance may be determined based at least in part on the determined state.","assignee":"Dexcom Inc","inventors":["Scott M. Belliveau","Naresh C. Bhavaraju","Jennifer Blackwell","Eric Cohen","Alexandra Elena CONSTANTIN","Basab Dattaray","Anna Leigh Davis","Rian DRAEGER","Arturo Garcia","John Michael Gray","Hari Hampapuram","Nathaniel David Heintzman","Lauren Hruby Jepson","Matthew Lawrence Johnson","Apurv Ullas Kamath","Katherine Yerre Koehler","Phil Mayou","Patrick Wile Mcbride","Michael Robert Mensinger","Sumitaka MIKAMI","Subrai Girish PAI","Andrew Attila Pal","Nicholas Polytaridis","Philip Thomas Pupa","Eli Reihman","Peter C. Simpson","Matthew T. Vogel","Tomas C. WALKER","Daniel Justin WEIDEBACK"],"publication_date":"2026-04-02","filing_date":"2026-03-16","priority_date":"2018-02-09","cpc_codes":["A","A61","A61B","A61B5/00","A61B5/145","A61B5/14532","A","A61","A61B","A61B5/00","A61B5/0002","A61B5/0015","A61B5/0022","A","A61","A61B","A61B5/00","A61B5/0002","A61B5/0015","A61B5/0024","A","A61","A61B","A61B5/00","A61B5/01","A","A61","A61B","A61B5/00","A61B5/02","A61B5/0205","A61B5/02055","A","A61","A61B","A61B5/00","A61B5/103","A61B5/11","A61B5/1118","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4836","A61B5/4839","A","A61","A61B","A61B5/00","A61B5/48","A61B5/486","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4866","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7221","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7275","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7282","A","A61","A61B","A61B5/00","A61B5/74","A61B5/742","A61B5/7435","A","A61","A61B","A61B5/00","A61B5/74","A61B5/746","A","A61","A61B","A61B5/00","A61B5/74","A61B5/7475","G","G06","G06N","G06N20/00","G","G06","G06N","G06N5/00","G06N5/04","G06N5/045","G","G16","G16H","G16H20/00","G16H20/10","G16H20/17","G","G16","G16H","G16H40/00","G16H40/60","G16H40/67","G","G16","G16H","G16H70/00","G16H70/20","A","A61","A61B","A61B2560/00","A61B2560/02","A61B2560/0242"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201971A1/en"},{"publication_number":"AU2026201963A1","title":"Concept for a defect center-based quantum computer based on a substrate of IVth group elements","abstract":"The invention relates to a quantum bit (QUB) having a quantum dot (NV), which may in particular be an NV center, and a nuclear quantum bit having at least one nuclear quantum dot, which is typically a nuclear spin afflicted isotope. These comprise a specific device for controlling a quantum dot (NV). Compounded therefrom, the invention comprises a quantum register of at least two quantum bits, a nuclear quantum register of at least two nuclear quantum bits, and a nucleus-electron quantum register of one quantum bit and one nuclear quantum bit, and a nucleus-electron-nucleus-electron quantum register of at least one quantum register and at least two nucleus-electron registers. A higher-level structure, a quantum bus, for transporting a quantum information and a quantum computer composed thereof are part of the invention. Also included in the invention are methods necessary to fabricate and operate the device. The invention consists primarily of the first overall assembly of all these devices and methods. Figure 21","assignee":"SaxonQ GmbH","inventors":["Bernd Burchard"],"publication_date":"2026-04-02","filing_date":"2026-03-16","priority_date":"2019-10-28","cpc_codes":["H","H10","H10D","H10D64/00","H10D64/20","H10D64/27","G","G06","G06N","G06N10/00","G06N10/40","H","H10","H10D","H10D48/00","H10D48/383","H10D48/3835","B","B82","B82Y","B82Y10/00"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201963A1/en"},{"publication_number":"KR20260044867A","title":"Method and Device for Personalized Notification Management Based on Feedback Analysis of User's Active Behavior","abstract":"본 발명은 사용자의 능동 활동에 대한 피드백 분석 기반의 맞춤형 알림 관리 방법 및 장치에 관한 것이다. 알림 수신부(110)가 어플리케이션(200)으로부터 알림 메시지(10)를 수신하면, 자연어 처리부(120)가 텍스트 데이터(11)를 분석하여 핵심 키워드(12) 및 문맥 정보(13)를 추출한다. 피드백 판별부(130)의 언어 패턴 분석 모듈(131)은 2인칭 대명사, 응답형 어미, 완료형 표현 등의 텍스트 패턴을 검출하고, 시간적 상관관계 분석 모듈(132)은 이벤트 발생 시점(20)과 알림 메시지(10) 수신 시각 간의 시간 차이를 이용하여 상관 계수(16)를 산출하며, 식별자 매칭 모듈(133)은 알림 메시지(10) 내에 사용자 고유 식별자(17)나 이전 질문 어휘의 포함 여부를 검출한다. 개인화 지수 산출부(140)는 상기 세 모듈의 결과를 종합하여 개인화 지수(14)를 산출하고, 우선순위 결정부(150)는 이를 기반으로 알림의 표시 우선순위를 결정하며, 알림 표시 제어부(160)는 알림 리스트(310)의 배치 순서 및 시각적 강조 영역(320)의 강조 효과를 제어한다. 또한, 학습 갱신부(170)가 사용자의 실제 반응 데이터를 기계 학습으로 알고리즘 가중치를 갱신하여 개인화 정확도를 점진적으로 향상시킨다. The present invention relates to a method and apparatus for managing customized notifications based on feedback analysis of a user's active activity. When a notification receiving unit (110) receives a notification message (10) from an application (200), a natural language processing unit (120) analyzes text data (11) to extract core keywords (12) and context information (13). A language pattern analysis module (131) of a feedback discrimination unit (130) detects text patterns such as second-person pronouns, response endings, and perfective expressions, a temporal correlation analysis module (132) calculates a correlation coefficient (16) using the time difference between the time of event occurrence (20) and the time of receiving the notification message (10), and an identifier matching module (133) detects whether the notification message (10) contains a unique user identifier (17) or previous question vocabulary. The personalization index calculation unit (140) calculates a personalization index (14) by combining the results of the three modules, the priority determination unit (150) determines the display priority of the notification based on this, and the notification display control unit (160) controls the arrangement order of the notification list (310) and the highlighting effect of the visual highlighting area (320). In addition, the learning update unit (170) gradually improves personalization accuracy by updating algorithm weights using machine learning based on the user's actual response data.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-04-02","filing_date":"2026-03-15","priority_date":"2026-03-15","cpc_codes":["H","H04","H04L","H04L67/00","H04L67/50","H04L67/55","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06N","G06N20/00","H","H04","H04L","H04L67/00","H04L67/2866","H04L67/30","H04L67/306","H","H04","H04L","H04L67/00","H04L67/50","H04L67/535"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260044867A/en"},{"publication_number":"AU2026201927A1","title":"Food contamination prediction device, inference device, machine learning device, food contamination prediction method, inference method, and machine learning method","abstract":"[Object] To provide a food contamination prediction device, an inference device, a machine learning device, a food contamination prediction method, an inference method, and a machine learning method that can easily predict the occurrence of a hazardous substance without directly inspecting the hazardous substance. [Solution] A food contamination prediction device (6) includes an information acquisition unit (600) that acquires environmental contamination indicator information including a detection status of an environmental contamination indicator different to a hazardous substance, and a prediction unit (601) configured to predict the occurrence status of the hazardous substance at the prediction point on the basis of the hazardous substance information, which is output when the environmental contamination indicator information acquired by the information acquisition unit (600) is input to a training model (12) learned by machine learning to learn a correlation between the environmental contamination indicator information including the detection status of the environmental contamination indicator at a training point and hazardous substance information including the occurrence status of the hazardous substance at the training point. [Selected drawing] FIG. 12","assignee":"Toyo Seikan Group Holdings Ltd","inventors":["Satoshi Furukawa","Hidehiko Kunimasa","Hiroshi Okamura","Suguru Tanabe"],"publication_date":"2026-04-02","filing_date":"2026-03-13","priority_date":"2021-11-02","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/04","G","G01","G01N","G01N33/00","G01N33/02","G","G06","G06N","G06N20/00"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201927A1/en"},{"publication_number":"MX2026003108A","title":"Support of vertical federated learning","abstract":"Various aspects of the present disclosure relate to a network entity of a wireless communication system, the network entity comprising at least one memory and at least one processor coupled with the at least one memory and configured to cause the network entity to receive an application layer request from a consumer entity for supporting a machine learning enabled application service, wherein the application layer request comprises a machine learning model identifier and/or an analytics event identifier, and determine a first requirement for providing vertical federated learning training for the machine learning enabled application service, based on the application layer request. Based on the first requirement, a second requirement is determined, wherein the second requirement comprises a data set requirement for the machine learning enabled service.","assignee":"Lenovo Singapore Pte Ltd","inventors":["Emmanouil Pateromichelakis","Dimitrios Karampatsis","Konstantinos Samdanis"],"publication_date":"2026-04-01","filing_date":"2026-03-12","priority_date":"2023-10-02","cpc_codes":["G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5061","G06F9/5072","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/502"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2026003108A/en"},{"publication_number":"MX2026003106A","title":"Notification handling for vertical federated learning enablement","abstract":"Various aspects of the present disclosure relate to a network entity for wireless communication with an AI/ML Service Registry, the AI/ML Service Registry being configured to manage a subscription for notifications related to a Federated Learning (FL) service, the network entity comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the network entity to: send a subscription request to the AI/ML Service Registry, wherein the subscription request comprises a request to receive information on at least one application layer event type; and receive, from the AI/ML Service Registry, information related to the at least one application layer event type.","assignee":"Lenovo Singapore Pte Ltd","inventors":["Emmanouil Pateromichelakis","Konstantinos Samdanis"],"publication_date":"2026-04-01","filing_date":"2026-03-12","priority_date":"2023-10-02","cpc_codes":["H","H04","H04L","H04L67/00","H04L67/50","H04L67/55","G","G06","G06N","G06N20/00"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2026003106A/en"},{"publication_number":"KR20260043091A","title":"Method implemented by computer, server apparatus, information processing system, program, and client terminal","abstract":"이동 단말기가 서버 장치와 통신할 수 없는 상태일 때에도, 이동 단말기가 디지털 어시스턴트로 기능하는 것을 가능하게 하는 기술을 제공한다. 사용자 단말기(200)가 사용자로부터 질의 A를 수신하면, 사용자 단말기(200)는 질의 A를 서버(100)에 송신한다. 서버(100)는 문법 A를 사용해서 질의 A의 의미를 해석한다. 서버(100)는 질의 A의 의미에 기초하여 질의 A에 대한 응답을 획득하고 이 응답을 사용자 단말기(200)에 송신한다. 서버(100)는 또한 질의 A를 사용자 단말기(200)에 송신한다. 즉, 서버(100)는 사용자 단말기(200)으로부터 수신된 질의를 해석하는 데 사용될 문법을 사용자 단말기(200)에 송신한다. It provides technology that enables a mobile terminal to function as a digital assistant even when the mobile terminal is unable to communicate with a server device. When the user terminal (200) receives query A from the user, the user terminal (200) transmits query A to the server (100). The server (100) interprets the meaning of query A using grammar A. Based on the meaning of query A, the server (100) obtains a response to query A and transmits this response to the user terminal (200). The server (100) also transmits query A to the user terminal (200). That is, the server (100) transmits to the user terminal (200) the grammar to be used to interpret the query received from the user terminal (200).","assignee":"사운드하운드, 인코포레이티드","inventors":["칼 스탈"],"publication_date":"2026-03-31","filing_date":"2026-03-24","priority_date":"2019-10-23","cpc_codes":["G","G06","G06F","G06F40/00","G06F40/20","G06F40/253","G","G06","G06F","G06F16/00","G06F16/20","G06F16/27","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3343","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3344","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/957","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G06F40/211","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G06F40/216","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G10","G10L","G10L15/00","G10L15/06","G10L15/063","G","G10","G10L","G10L15/00","G10L15/06","G10L15/065","G10L15/07","G","G10","G10L","G10L15/00","G10L15/08","G10L15/18","G10L15/183","G","G10","G10L","G10L15/00","G10L15/22"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260043091A/en"},{"publication_number":"KR20260042450A","title":"Method of screening normal tissue using artificial intelligence and electronic device performing the same","abstract":"본 개시는 인공지능을 이용하여 정상 조직을 스크리닝하는 방법 및 이를 수행하는 전자 장치에 관한 것이다. 본 개시의 일 실시예에 따른 방법은, 조직 샘플에 대응하는 복수의 패치들을 획득하는 단계, 복수의 패치들을 입력으로 하는 제1 인공지능 모델을 이용하여, 복수의 신호 값들을 추출하는 단계, 복수의 신호 값들 및 복수의 패치들에 대응하는 위치 값들을 입력으로 하는 제2 인공지능 모델을 이용하여, 복수의 임베딩 벡터들을 획득하는 단계, 복수의 임베딩 벡터들을 입력으로 하는 제3 인공지능 모델을 이용하여, 조직 샘플 내에 병변이 있는지 결정하는 단계, 및 조직 샘플 내에 병변이 없는 것으로 결정한 것에 기초하여, 조직 샘플이 정상인 것으로 결정하는 단계를 포함하고, 제1 인공지능 모델은, 복수의 패치들에 기초하여 복수의 신호 값들을 출력하는 인코더, 및 복수의 신호 값들에 기초하여 복수의 패치들의 기질 부분과 상피 부분이 구분된 이미지를 출력하는 디코더를 포함할 수 있다. The present disclosure relates to a method for screening normal tissue using artificial intelligence and an electronic device for performing the same. A method according to one embodiment of the present disclosure comprises the steps of: acquiring a plurality of patches corresponding to a tissue sample; extracting a plurality of signal values using a first artificial intelligence model that takes the plurality of patches as input; acquiring a plurality of embedding vectors using a second artificial intelligence model that takes the plurality of signal values and position values corresponding to the plurality of patches as input; determining whether there is a lesion in the tissue sample using a third artificial intelligence model that takes the plurality of embedding vectors as input; and determining that the tissue sample is normal based on the determination that there is no lesion in the tissue sample. The first artificial intelligence model may include an encoder that outputs a plurality of signal values based on the plurality of patches, and a decoder that outputs an image in which the stromal and epithelial portions of the plurality of patches are distinguished based on the plurality of signal values.","assignee":"주식회사 프리닥터","inventors":["윤지섭"],"publication_date":"2026-03-31","filing_date":"2026-03-19","priority_date":"2024-08-09","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06V","G06V10/00","G06V10/20","G06V10/25","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/60","G06V20/69","G","G16","G16H","G16H30/00","G16H30/40","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/50","G","G16","G16H","G16H50/00","G16H50/70","G","G06","G06V","G06V2201/00","G06V2201/03"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260042450A/en"},{"publication_number":"KR20260043610A","title":"Method and System for Pre-compensating Doppler Shift in Satellite Communication for Drones Using Flight Plan Data","abstract":"본 발명은 비행 계획 데이터(130)를 이용한 드론(100)의 위성 통신 도플러 시프트 선보정 방법 및 도플러 시프트 선보정 시스템(10)에 관한 것이다. 본 발명은 비행 제어 장치(110) 또는 미션 플래너(120)로부터 웨이포인트(131), 이동 속도 정보(132), 임무 유형 정보(133), 선회 반경 정보(134), 가속도 변화 패턴(135)을 포함하는 비행 계획 데이터(130)를 획득하고, 위성(200)의 궤도 정보(210)와 함께 AI 모델(310)에 입력하여 미래 각 시점에서의 드론(100)과 위성(200) 간 상대 속도 벡터 및 도플러 시프트를 사전 예측함으로써, 통신 모듈(400)의 반송파 주파수 조정부(410)가 실제 기동 이전에 선제적으로 반송파 주파수를 최적값으로 조정한다. 이를 통해 드론(100)의 급격한 기동 구간에서도 위성(200) 통신 링크의 동기 이탈 없이 안정적인 통신을 유지할 수 있으며, GPS 수신부(500)의 실시간 위치 정보(510)를 오차 학습부(315)에 피드백하여 AI 모델(310)이 위성(200)의 궤도 정보 오차 패턴을 지속적으로 학습 및 갱신하는 자기 개선 기능을 제공한다. The present invention relates to a method for pre-correcting satellite communication Doppler shift of a drone (100) using flight plan data (130) and a Doppler shift pre-correcting system (10). The present invention obtains flight plan data (130) including waypoints (131), movement speed information (132), mission type information (133), turning radius information (134), and acceleration change pattern (135) from a flight control device (110) or a mission planner (120), and inputs it into an AI model (310) together with orbit information (210) of a satellite (200) to predict in advance the relative velocity vector and Doppler shift between the drone (100) and the satellite (200) at each future point in time, thereby allowing the carrier frequency adjustment unit (410) of the communication module (400) to pre-adjust the carrier frequency to an optimal value before actual operation. Through this, stable communication can be maintained without synchronization loss of the satellite (200) communication link even during rapid maneuvering of the drone (100), and a self-improvement function is provided in which the AI model (310) continuously learns and updates the orbital information error pattern of the satellite (200) by feeding back the real-time location information (510) of the GPS receiver (500) to the error learning unit (315).","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-03-31","filing_date":"2026-03-14","priority_date":"2026-03-14","cpc_codes":["H","H04","H04B","H04B7/00","H04B7/01","B","B64","B64C","B64C39/00","B64C39/02","B64C39/024","G","G05","G05D","G05D1/00","G05D1/20","G05D1/24","G05D1/247","G05D1/248","G","G05","G05D","G05D1/00","G05D1/60","G05D1/644","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","H","H04","H04W","H04W56/00","H04W56/0035","H","H04","H04W","H04W84/00","H04W84/02","H04W84/04","H04W84/06","B","B64","B64U","B64U2101/00","G","G05","G05D","G05D2109/00","G05D2109/20"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260043610A/en"},{"publication_number":"KR20260043081A","title":"Blockchain-powered Contribution Assessment and Token Incentive System for Generative AI Refinement","abstract":"본 발명은 블록체인 기반 생성형 AI 응답 교정 기여도 측정 및 토큰 보상 시스템에 관한 것으로, 사용자 단말기(100) 및 생성형 AI 서버(200)와 네트워크(700)로 연결된 보상 시스템 서버(300)를 포함한다. 상기 보상 시스템 서버(300)는 피드백 수집 모듈(310), 교정 유효성 검증 엔진(320), 기여도 산출부(330), 블록체인 기록부(340) 및 스마트 계약 기반 보상부(350)를 포함한다. 피드백 수집 모듈(310)은 생성형 AI 서버(200)의 원본 응답과 사용자의 수정 응답을 쌍으로 수집하고, 교정 유효성 검증 엔진(320)은 검증용 AI 모델(600) 및 복수의 사용자 교차 검증을 통해 교정의 유효성을 판별하며, 기여도 산출부(330)는 오류 유형별 가중치를 적용하여 기여 점수를 산출한다. 블록체인 기록부(340)는 검증된 기여 이력을 블록체인 네트워크(400)의 분산 원장(410)에 불변 기록하고, 스마트 계약 기반 보상부(350)는 스마트 계약부(420)을 통해 사용자의 가상화폐 지갑(500)으로 토큰을 자동 전송한다. 본 발명에 의하면 교정 데이터의 무결성 보장, 공정한 보상 자동화, 및 생성형 AI 품질의 지속적 향상이 가능하다. The present invention relates to a blockchain-based generative AI response correction contribution measurement and token reward system, comprising a reward system server (300) connected to a user terminal (100), a generative AI server (200), and a network (700). The reward system server (300) comprises a feedback collection module (310), a correction validity verification engine (320), a contribution calculation unit (330), a blockchain record unit (340), and a smart contract-based reward unit (350). The feedback collection module (310) collects the original response of the generative AI server (200) and the user's modified response in pairs, the correction validity verification engine (320) determines the validity of the correction through a verification AI model (600) and a plurality of user cross-verification, and the contribution calculation unit (330) calculates a contribution score by applying weights according to error types. The blockchain record book (340) immutably records the verified contribution history in the distributed ledger (410) of the blockchain network (400), and the smart contract-based reward book (350) automatically transfers tokens to the user's virtual currency wallet (500) through the smart contract book (420). According to the present invention, it is possible to ensure the integrity of correction data, automate fair rewards, and continuously improve the quality of generative AI.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-03-31","filing_date":"2026-03-13","priority_date":"2026-03-13","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/27","G","G06","G06F","G06F16/00","G06F16/20","G06F16/23","G06F16/2365","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/242","G06F16/243","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/248","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","H","H04","H04L","H04L9/00","H04L9/50"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260043081A/en"},{"publication_number":"KR20260043075A","title":"System and Method for Automatically Issuing Blockchain-Based Carbon Credit Tokens via Counterfactual Trajectory Inference and Distributed Oracle Consensus for Measurement and Verification of Individual Carbon Avoidance","abstract":"본 발명은 사용자의 보행 이동이 자동차 또는 택시 이용을 대체하였음을 인 공지능 기반 반사실적 경로 추론 모듈이 확률적으로 추론하고, 이를 근거로 산출한 탄소 회피량을 복수의 독립 오라클 노드로 구성된 분산 오라클 합의 네트워크가 검 증하며, 블록체인 스마트 계약이 탄소 배출권 토큰을 자동 발행하는 시스템 및 방 법에 관한 것이다. 반사실적 경로 추론은 개인의 과거 이동 이력, 실시간 교통 환 경 데이터, 베이지안 사후 확률 추정을 결합하여 수행되며, 분산 오라클 합의는 비 잔틴 내결함성 프로토콜로 데이터 조작을 방지한다. 또한 영지식 증명(ZKP) 기법을 적용하여 사용자의 위치 프라이버시를 보호하면서 탄소 회피량의 진실성을 수학적 으로 증명하며, 공인 탄소 배출권 등록소와의 자동 연동으로 이중계산을 방지한다. The present invention relates to a system and method in which an AI-based counterfactual path inference module probabilistically infers that a user's walking movement has replaced the use of a car or taxi, a distributed oracle consensus network composed of multiple independent oracle nodes verifies the carbon avoidance amount calculated based on this, and a blockchain smart contract automatically issues carbon emission rights tokens. Counterfactual path inference is performed by combining an individual's past movement history, real-time traffic environment data, and Bayesian posterior probability estimation, and the distributed oracle consensus prevents data manipulation using a Byzantine fault tolerance protocol. Furthermore, by applying Zero-Knowledge Proof (ZKP) techniques, the truthfulness of the carbon avoidance amount is mathematically proven while protecting the user's location privacy, and double counting is prevented through automatic linkage with an accredited carbon emission rights registry.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-03-31","filing_date":"2026-03-12","priority_date":"2026-03-12","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q40/00","G06Q40/04","G06Q40/049","G","G06","G06Q","G06Q50/00","G06Q50/40","H","H04","H04L","H04L9/00","H04L9/50"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260043075A/en"},{"publication_number":"KR20260042150A","title":"Operating method for electronic apparatus for providing information and electronic apparatus supporting thereof","abstract":"본 개시에 따르면, 전자 장치가 정보를 제공하는 방법에 있어서, 상기 전자 장치와 관련된 서비스 상에서, 반려 동물의 케어(care)를 위한 정보를 제공하는 반려 동물 서비스 페이지를 선택한 사용자의 입력에 대응하여 상기 반려 동물 서비스 페이지를 제공하는 단계; 상기 반려 동물 서비스 페이지를 통해 상기 사용자의 반려 동물의 제1 건강 상태 정보를 획득하는 단계; 기 설정된 모델에 기반하여, 상기 제1 건강 상태 정보에 대응하는 상기 반려 동물을 위한 제1 추천 아이템에 대한 정보를 확인하는 단계; 및 상기 제1 추천 아이템에 대한 정보를 제공하는 단계를 포함하는 방법이 개시된다. According to the present disclosure, a method for an electronic device to provide information is disclosed, comprising: providing a pet service page in response to input by a user who selects a pet service page that provides information for the care of a pet on a service associated with the electronic device; obtaining first health status information of the user's pet through the pet service page; confirming information about a first recommended item for the pet corresponding to the first health status information based on a pre-set model; and providing information about the first recommended item.","assignee":"쿠팡 주식회사","inventors":["전선규","김은미","최성은","오혜진","이충림"],"publication_date":"2026-03-30","filing_date":"2026-03-23","priority_date":"2023-06-13","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0631","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q30/00","G","G06","G06Q","G06Q30/00","G06Q30/018","G","G06","G06Q","G06Q30/00","G06Q30/06","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/22","G","G16","G16H","G16H20/00","G","G16","G16H","G16H40/00","G16H40/20","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/70","G","G16","G16H","G16H80/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260042150A/en"},{"publication_number":"KR20260042127A","title":"Apparatus and method for generating a depth map using a volumetric feature","abstract":"본 개시는 볼류메트릭 피처를 이용하여 깊이 맵을 생성하는 장치 및 방법에 관한 것이다. 본 개시의 일 실시 예에 따른 방법은, 서라운드 뷰 이미지에 포함된 기본 이미지를 인코딩 및 후처리하여 기본 이미지에 대한 단일 피처맵을 생성하고, 단일 피처맵을 깊이 정보와 함께 인코딩한 후, 인코딩 결과를 3차원 공간으로 투영함으로써 볼류메트릭 피처를 생성할 수 있다. 본 방법에서는 깊이 디코더를 이용하여 볼류메트릭 피처를 디코딩함으로써, 서라운드 뷰 이미지의 깊이 맵을 생성할 수 있다. The present disclosure relates to an apparatus and method for generating a depth map using volumetric features. A method according to one embodiment of the present disclosure can generate a volumetric feature by encoding and post-processing a base image included in a surround view image to generate a single feature map for the base image, encoding the single feature map together with depth information, and then projecting the encoding result into a three-dimensional space. In the present method, a depth map of a surround view image can be generated by decoding the volumetric feature using a depth decoder.","assignee":"포티투닷 주식회사","inventors":["김중희","푸옥 응우옌 티엔","정성균","허준화"],"publication_date":"2026-03-30","filing_date":"2026-03-11","priority_date":"2022-07-29","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/50","G06T7/55","G06T7/593","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T3/00","G06T3/40","G","G06","G06T","G06T7/00","G06T7/70","H","H04","H04N","H04N13/00","H04N13/10","H04N13/106","H04N13/161","H","H04","H04N","H04N13/00","H04N13/20","H04N13/204","H04N13/207","H","H04","H04N","H04N13/00","H04N13/30","H04N13/388","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30244"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260042127A/en"},{"publication_number":"KR20260041756A","title":"Method and apparatus for video encoding and video decoding based on neural network","abstract":"비디오의 복호화 방법, 복호화 장치, 부호화 방법 및 부호화 장치가 개시된다. 생성-부호기 및 생성-복호기를 포함하는 비디오 생성 네트워크에 의해 가상 프레임이 생성된다. 가상 프레임은 대상에 대한 인터 예측에서 참조 프레임으로서 사용된다. 또한, 복수의 비디오 생성 네트워크들 중 인터 예측을 위한 비디오 생성 네트워크가 선택될 수 있고, 선택된 비디오 생성 네트워크를 사용하는 인터 예측이 수행될 수 있다. A method for decoding video, a decoding device, an encoding method, and an encoding device are disclosed. A virtual frame is generated by a video generation network comprising a generator-encoder and a generator-decoder. The virtual frame is used as a reference frame in inter prediction for a target. Additionally, among a plurality of video generation networks, a video generation network for inter prediction may be selected, and inter prediction using the selected video generation network may be performed.","assignee":"한국전자통신연구원; 이화여자대학교 산학협력단","inventors":["조승현","김연희","석진욱","이주영","임웅","김종호","이대열","정세윤","김휘용","최진수","강제원","김나영"],"publication_date":"2026-03-27","filing_date":"2026-03-16","priority_date":"2018-02-08","cpc_codes":["H","H04","H04N","H04N19/00","H04N19/60","H04N19/61","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/103","H04N19/109","H","H04","H04N","H04N19/00","H04N19/50","H04N19/503","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/103","H04N19/11","H","H04","H04N","H04N19/00","H04N19/10","H04N19/169","H04N19/17","H04N19/172","H","H04","H04N","H04N19/00","H04N19/10","H04N19/169","H04N19/17","H04N19/176","H","H04","H04N","H04N19/00","H04N19/10","H04N19/169","H04N19/184","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260041756A/en"},{"publication_number":"KR20260041741A","title":"Version Hallucination Blocking System via Spatiotemporal Conflict Graph and LLM Internal Gating Control, and Uncertainty-Based Self-Correction Method","abstract":"본 발명은 대화형 인공지능의 시공간 환각 및 생성적 오류를 차단하는 구동 방법 및 시스템에 관한 것이다. 본 발명은 지식 그래프의 상충 관계를 기반으로 LLM 내부 파라미터를 하드 차단하거나 사용자의 창조적 우회 의도에 따라 조건부 소프트 라우팅을 수행하여 버전 혼합 오류를 시스템 레벨에서 차단한다. 또한, 확률적 샘플링 과정에서 불확실성 지표 급증 시 서브워드를 임시 버퍼링하여 생성적 충돌을 진단하고 실시간으로 자가 교정한다. 나아가, 생성된 응답의 근거가 되는 시간 메타데이터를 포함하는 시간적 책임 레이어(Temporal Accountability Layer) 레코드를 생성하여 컴플라이언스 대응력을 극대화한다. 이를 통해 인공지능 시스템의 기술적 신뢰도와 B2B 환경에서의 사용자 경험을 동시에 향상시킨다. The present invention relates to an operating method and system for blocking spatiotemporal illusions and generative errors in conversational artificial intelligence. The invention blocks version mixing errors at the system level by hard-blocking internal LLM parameters based on trade-offs in a knowledge graph or by performing conditional soft routing according to the user's creative bypass intentions. Furthermore, it diagnoses generative conflicts and performs real-time self-correction by temporarily buffering subwords when uncertainty indicators surge during the probabilistic sampling process. Moreover, it maximizes compliance capabilities by creating Temporal Accountability Layer records that include temporal metadata serving as the basis for the generated response. Through this, it simultaneously enhances the technical reliability of the artificial intelligence system and the user experience in B2B environments.","assignee":"이우철","inventors":["이우철"],"publication_date":"2026-03-27","filing_date":"2026-03-10","priority_date":"2026-03-10","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9032","G06F16/90332","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G06F16/9024","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9038","G","G06","G06F","G06F16/00","G06F16/90","G06F16/907","G","G06","G06N","G06N20/00","G","G06","G06F","G06F2201/00","G06F2201/81"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260041741A/en"},{"publication_number":"KR20260041734A","title":"Method and system for generating unit cost table of advertisement including a plurality of unit costs of advertisement based on predicted unit cost of advertisement with respect to advertisement region","abstract":"광고 네트워크가 과거에 광고 영역을 구매한 광고 단가를 포함하는 광고 영역에 대한 이력 정보에 기반하여, 광고 네트워크가 광고 영역을 구매할 것으로 예상되는 광고 단가를 예측하고, 예측된 광고 단가에 기반하여 결정된 복수의 광고 단가들을 포함하는 광고 단가 테이블을 생성하는, 광고 단가 테이블의 생성 방법이 제공된다. A method for generating an ad price table is provided, which predicts the ad price at which an ad network is expected to purchase an ad area based on historical information regarding an ad area including the ad price at which the ad network previously purchased the ad area, and generates an ad price table including a plurality of ad prices determined based on the predicted ad price.","assignee":"네이버웹툰 유한회사","inventors":["노인우","오혜지"],"publication_date":"2026-03-27","filing_date":"2026-03-04","priority_date":"2022-09-06","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0273","G06Q30/0275","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0273","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0206","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0242"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260041734A/en"},{"publication_number":"CN121743321A","title":"Data circulation risk identification method and system based on multiple vertical large model cognitive games","abstract":"本发明公开了一种基于多个垂类大模型认知博弈的数据流通风险识别方法及系统，属于人工智能技术领域。该方法包括：构建包含多源数据的语料库，经处理后建立支持混合检索的向量知识库；设计融合通用维度与动态场景化维度的合规评价体系；利用领域数据训练得到数据合规领域的垂类大模型；基于智能体检索增强生成与思维链技术，驱动多个垂类大模型进行结构化认知博弈与迭代收敛，形成共识性风险清单；最后通过案例匹配验证生成风险识别报告。本发明通过多源知识融合和多个垂类大模型认知博弈机制，抑制了大模型输出的随机性，解决了人工审核效率低与通用大模型专业性不足的问题，实现了高效、可靠的数据流通风险识别。 This invention discloses a method and system for identifying data circulation risks based on cognitive game theory using multiple large-scale vertical models, belonging to the field of artificial intelligence technology. The method includes: constructing a corpus containing multi-source data; processing this corpus to establish a vector knowledge base supporting hybrid retrieval; designing a compliance evaluation system integrating general dimensions and dynamic, scenario-based dimensions; training large-scale vertical models in the data compliance field using domain data; driving multiple large-scale vertical models to perform structured cognitive game theory and iterative convergence based on intelligent agent retrieval enhancement generation and thought chain technology to form a consensus-based risk list; and finally generating a risk identification report through case matching verification. This invention, through multi-source knowledge fusion and the cognitive game theory mechanism of multiple large-scale vertical models, suppresses the randomness of the large-scale model output, solves the problems of low efficiency in manual review and insufficient professionalism of general large-scale models, and achieves efficient and reliable identification of data circulation risks.","assignee":"Shanghai Pudong New Area Big Data Center","inventors":["赵琳","黄得志","曾驰斌","陈媛"],"publication_date":"2026-03-27","filing_date":"2026-02-27","priority_date":"2025-12-29","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/21","G06F16/215","G","G06","G06F","G06F16/00","G06F16/20","G06F16/22","G06F16/2228","G06F16/2237","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3347","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/335","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06N","G06N5/00","G06N5/04","G06N5/042"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121743321A/en"},{"publication_number":"CN121580341B","title":"Rotating machinery fault diagnosis method and system based on cross-modal time-frequency sensing","abstract":"本发明涉及机械故障诊断技术领域，尤其是指一种基于跨模态时频感知的旋转机械故障诊断方法及系统，设计了时域、频域和时频域感知嵌入分支，能够自适应处理来自不同设备的跨模态的时域、频域和时频域信号，实现多模态信号的有效融合，提升故障诊断的精度和鲁棒性。针对时域和频域的私有特征属性和时频域的公有特征属性，本发明分别对时域、频域和时频域的嵌入特征设计对应的Transformer分支，并利用时频汇聚多头交叉注意力模块，以时域和频域的局部特征表示，聚合全局时频域特征表示，充分挖掘时域与频域之间的共性和差异，进一步提升故障诊断的准确性和鲁棒性。 This invention relates to the field of mechanical fault diagnosis technology, and in particular to a method and system for diagnosing rotating machinery faults based on cross-modal time-frequency sensing. It designs time-domain, frequency-domain, and time-frequency-domain sensing embedding branches, enabling adaptive processing of cross-modal time-domain, frequency-domain, and time-frequency-domain signals from different devices, achieving effective fusion of multi-modal signals, and improving the accuracy and robustness of fault diagnosis. For the private feature attributes in the time and frequency domains and the common feature attributes in the time-frequency domain, this invention designs corresponding Transformer branches for the embedded features in the time, frequency, and time-frequency domains respectively, and utilizes a time-frequency convergence multi-head cross-attention module to aggregate global time-frequency domain feature representations using local feature representations in the time and frequency domains, fully exploring the commonalities and differences between the time and frequency domains, further improving the accuracy and robustness of fault diagnosis.","assignee":"Suzhou University","inventors":["陈良","陈启通"],"publication_date":"2026-03-27","filing_date":"2026-01-29","priority_date":"2026-01-29","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/254","G06F18/256","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06F","G06F2218/00","G06F2218/08","G","G06","G06F","G06F2218/00","G06F2218/12"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121580341B/en"},{"publication_number":"CN121581603B","title":"Driving mechanism analysis method for non-uniform evolution of reservoir group scheduling on flood","abstract":"The invention discloses a driving mechanism analysis method for non-uniform evolution of reservoir group scheduling on floods, which comprises the steps of obtaining a smooth warehousing flood sequence by using a physical manifold constraint denoising model of topological coupling, constructing dynamic topological features for representing propagation time lag and sensitivity by using a graph-nerve ordinary differential equation based on a hydraulic propagation mechanism, constructing a structural causal model comprising scheduling capability indexes, exogenous hydrologic driving and topological features, fitting a nonlinear dependency relationship by using a causal generalized additive model, executing inverse fact inference, calculating a local causal driving index and a global causal cumulative effect index by using an intervention operator, and generating an optimized scheduling strategy for inhibiting variation based on the causal index. The invention can realize systematic analysis from data physical restoration to causal mechanism decoupling, accurately quantizes the driving contribution of scheduling behavior to flood non-uniformity, and provides decision support for river basin scientific flood control.","assignee":"Hohai University HHU","inventors":["王斌","钟平安","樊宇堃","朱非林","许成婧","徐斌","钱心缘","陈娟","本梦雪","万新宇"],"publication_date":"2026-03-27","filing_date":"2026-01-28","priority_date":"2026-01-28","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06312","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/23","G06F18/232","G06F18/2323","G","G06","G06N","G06N10/00","G06N10/60","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06N","G06N5/00","G06N5/04","G06N5/045","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26","G","G06","G06F","G06F2123/00","G06F2123/02","Y","Y02","Y02A","Y02A10/00","Y02A10/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121581603B/en"},{"publication_number":"CN121580087B","title":"Trend fault prediction method based on dynamic mode and threshold cooperation","abstract":"The invention relates to the technical field of industrial equipment state monitoring and fault diagnosis, in particular to a trend fault prediction method based on dynamic mode and threshold cooperation. According to the invention, a theoretical prediction interval which dynamically changes along with load is generated in real time by establishing nonlinear mapping between working conditions and key parameters, parameter drift interference caused by working condition fluctuation is effectively eliminated, a real-time health baseline of maintenance record quantification equipment is combined, the width of an early warning threshold is adjusted in a cooperative mode according to feature similarity and health level, self-adaptive monitoring of different aging stages of a whole life cycle is realized, multi-dimensional feature vector fusion physical field information is utilized for pattern matching, deviation severity and morphological similarity are comprehensively estimated through fuzzy reasoning, anomalies are locked in advance according to high feature fitness when the numerical value does not severely exceed the limit, early weak symptoms are accurately captured, an abnormal source is output, and the diagnosis accuracy under the variable working conditions is remarkably improved.","assignee":"Shenneng Smart Energy Technology Co ltd","inventors":["郑冉阳","徐晨博","马钧隆","梁山恒","石磊"],"publication_date":"2026-03-27","filing_date":"2026-01-27","priority_date":"2026-01-27","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G01","G01D","G01D21/00","G01D21/02","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2431","G","G06","G06F","G06F18/00","G06F18/20","G06F18/27","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","Y","Y04","Y04S","Y04S10/00","Y04S10/50"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121580087B/en"},{"publication_number":"CN121561170B","title":"Electronic archive management method and system based on knowledge graph","abstract":"本发明涉及电子档案管理技术领域，具体为基于知识图谱的电子档案管理方法及系统，其包括提取文书字段信息并绑定生成时间，构建语义片段并建立编号索引，筛查片段关联关系生成一致性值，更新图谱节点结构并合并表达内容，依据部门归属划分语义管理组并生成唯一标识。本发明通过对文书标题、编号、部门、时间与关键词进行字段联动，构建具备时序与归属属性的语义单元，依据关键词重合与顺序逻辑组合片段，生成结构连续的图谱节点并替换原有路径结构，结合部门与语义特征划分管理单元并分配唯一标识，完成档案信息聚合、节点结构更新与语义单元管理。 This invention relates to the field of electronic records management technology, specifically to a knowledge graph-based electronic records management method and system. It includes extracting document field information and binding it to the generation time, constructing semantic fragments and establishing a numbered index, screening fragment relationships to generate consistency values, updating the knowledge graph node structure and merging expressed content, and dividing semantic management groups according to department affiliation and generating unique identifiers. This invention constructs semantic units with temporal and affiliation attributes by linking document titles, numbers, departments, times, and keywords. Based on keyword overlap and sequential logical combination of fragments, it generates structurally continuous knowledge graph nodes and replaces the original path structure. Combining department and semantic features, it divides management units and assigns unique identifiers, thus completing the aggregation of archival information, updating of node structures, and management of semantic units.","assignee":"Tongluo Technology Co ltd","inventors":["肖宇","王志武","裴伟","冯德明","梁文佳","陈炳应"],"publication_date":"2026-03-27","filing_date":"2026-01-22","priority_date":"2026-01-22","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/93","G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G06F16/9024","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121561170B/en"},{"publication_number":"CN121543747B","title":"Answer generation method, device, electronic equipment and medium","abstract":"The application discloses an answer generation method, an answer generation device, electronic equipment and a medium, which relate to the technical field of computers, wherein the method comprises the steps of obtaining a firmware inquiry problem; and inputting the at least one target firmware, the at least one target firmware set and the firmware query question into a first pre-built reasoning model to generate an answer corresponding to the firmware query question. According to the application, the basis is provided for the first reasoning model through the firmware and the firmware set which are related with the firmware inquiry problem, so that the generated answer can be related with the semantic matching and execution invoking logic of the firmware inquiry problem, and the answer generation quality is improved.","assignee":"Inspur Jinan data Technology Co ltd","inventors":["郭立民","郭涛","冯振","孔维亭","刘元松","张霄炜","杜海超"],"publication_date":"2026-03-27","filing_date":"2026-01-21","priority_date":"2026-01-21","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06F","G06F11/00","G06F11/07","G06F11/0703","G06F11/079","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G06F16/353","G","G06","G06F","G06F40/00","G06F40/20","G06F40/253","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/455","G06F9/45533","G06F9/45558","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/455","G06F9/45533","G06F9/45558","G06F2009/45591"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121543747B/en"},{"publication_number":"CN121542864B","title":"Big language model teaching quality assessment method based on Brumu target classification system","abstract":"本发明属于自然语言处理技术领域，具体涉及一种基于布鲁姆目标分类体系的大语言模型教学质量评估方法，包括：根据布鲁姆目标分类体系构建结构化的提示词模板；通过提示词模板驱动大语言模型协同生成符合指定格式和分布要求的教学质量评估数据集；ChatGLM3预训练模型主要由28层基于Transformer网络的GLM模块构成，将部分GLM模块的多层感知机替换为混合专家网络，在ChatGLM3预训练模型的每层GLM模块前加入可训练的前缀编码，在词嵌入层之后、GLM模块之前加入Mamba状态空间模型，得到教学质量评估大语言模型，训练后用于教学质量评估。本发明可提升教学质量评估的准确性和针对性。 This invention belongs to the field of natural language processing technology, specifically relating to a teaching quality assessment method based on Bloom's objective classification system using a large language model. The method includes: constructing structured prompt word templates based on Bloom's objective classification system; using the prompt word templates to drive the large language model to collaboratively generate a teaching quality assessment dataset that conforms to specified format and distribution requirements; the ChatGLM3 pre-trained model mainly consists of 28 layers of GLM modules based on Transformer networks, replacing the multilayer perceptrons in some GLM modules with hybrid expert networks, adding trainable prefix codes before each GLM module layer in the ChatGLM3 pre-trained model, and adding a Mamba state-space model after the word embedding layer and before the GLM modules to obtain a large language model for teaching quality assessment, which is then trained and used for teaching quality assessment. This invention can improve the accuracy and relevance of teaching quality assessment.","assignee":"Jiangxi Agricultural University","inventors":["易文龙","黄抒"],"publication_date":"2026-03-27","filing_date":"2026-01-20","priority_date":"2026-01-20","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06N","G06N5/00","G06N5/04","G06N5/043","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/20"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121542864B/en"},{"publication_number":"CN121545186B","title":"A Video Pedestrian Re-identification Method and System Based on Riemannian Manifold Second-Order Relation Modeling","abstract":"The invention discloses a video pedestrian re-identification method and system based on Riemann manifold second-order relation modeling, and relates to the technical field of video pedestrian re-identification. Aiming at the problems that the prior method relies on first-order features and is difficult to capture high-order space-time correlation, the method comprises the steps of preprocessing video data, generating block feature representations, encoding the block feature representations into frame-level features, calculating inter-channel covariance matrixes by splicing the block features, optimizing the inter-channel covariance matrixes into positive definite symmetric matrixes, inputting a Rieman manifold network to execute multi-round bilinear mapping, nonlinear activation and cross-round feature fusion after a mask matrix is obtained through a random mask, and finally projecting the fusion features into a cut space to finish retrieval matching. According to the invention, the high-order associated information is mined through the second-order relation modeling, and the robustness and the recognition accuracy of the model to complex scenes are improved by combining the Riemann manifold geometric characteristics and the feature fusion mechanism, so that the method is suitable for the fields of intelligent security, video monitoring and the like.","assignee":"Jiangnan University","inventors":["徐天阳","李民志","于轩","周恒�","朱学峰","吴小俊","许亚骏"],"publication_date":"2026-03-27","filing_date":"2026-01-19","priority_date":"2026-01-19","cpc_codes":["G","G06","G06V","G06V40/00","G06V40/10","G","G06","G06F","G06F16/00","G06F16/70","G06F16/78","G06F16/783","G06F16/7837","G06F16/784","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06V","G06V10/00","G06V10/20","G","G06","G06V","G06V10/00","G06V10/40","G06V10/42","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/7715","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121545186B/en"},{"publication_number":"CN121545130B","title":"Construction Method and Application of Road Surface Type Recognition Model Based on Improved MobileNetV4","abstract":"本申请提出了一种基于改进MobileNetV4的路面类型识别模型的构建方法及应用，包括以下步骤：获取多张包含标注有路面类型的路面图像作为训练样本集合；基于MobileNetV4架构构建路面类型识别架构，并以训练样本集合对路面类型识别架构进行训练得到路面类型识别模型，其中，所述路面类型识别架构包括融合卷积瓶颈网络、通用倒置瓶颈网络以及分类器。本方案引入通道分组机制和深度可分离卷积，在极低的参数量和计算开销下，精准地捕获了对路面分类至关重要的细粒度纹理特征，并有效抑制了环境噪声的干扰，实现对路面类型的精准识别。 This application proposes a method for constructing and applying a road surface type recognition model based on an improved MobileNetV4, comprising the following steps: acquiring multiple road surface images labeled with road surface types as a training sample set; constructing a road surface type recognition architecture based on the MobileNetV4 architecture, and training the road surface type recognition architecture with the training sample set to obtain a road surface type recognition model, wherein the road surface type recognition architecture includes a fused convolutional bottleneck network, a general inverted bottleneck network, and a classifier. This scheme introduces a channel grouping mechanism and depthwise separable convolution, accurately capturing fine-grained texture features crucial for road surface classification with extremely low parameter count and computational overhead, and effectively suppressing environmental noise interference, thus achieving accurate road surface type recognition.","assignee":"Zhejiang Sci Tech University ZSTU","inventors":["王骏骋","李志�","倪喆源"],"publication_date":"2026-03-27","filing_date":"2026-01-16","priority_date":"2026-01-16","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/50","G06V20/56","G06V20/588","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","Y","Y02","Y02T","Y02T10/00","Y02T10/10","Y02T10/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121545130B/en"},{"publication_number":"AU2026201724A1","title":"Single pane of glass mobile application including erp agnostic realtime data mesh with data change capture","abstract":"System and methods are provided for real-time data integration, analysis, and notification within a Mobile App system. Embodiments include initializing a data layer and preprocessing phase, leveraging distributed database strategies to store structured and unstructured data. The Real-Time Data Mesh (RTDM) continuously draws data from various platforms, employing signal processing methods and machine learning techniques for noise removal and data feature extraction. The Advanced Analytics and Machine Learning (AAML) engine processes data, while decision constructs derive suitable actions. The Push Notification Service activates based on an Event-Driven Architecture (EDA) or a Publish-Subscribe (Pub- Sub) system, delivering customized notifications. Technical precision ensures timely and pertinent information reaches end-users. The system optimizes operations through adaptive feedback mechanisms and secures data through encryption. 20 26 20 17 24 06 M ar 2 02 6 2 0 2 6 2 0 1 7 2 4 0 6 M a r 2 0 2 6","assignee":"Ingram Micro Inc","inventors":["Sanjib Sahoo"],"publication_date":"2026-03-26","filing_date":"2026-03-06","priority_date":"2024-01-26","cpc_codes":["H","H04","H04L","H04L67/00","H04L67/50","H04L67/55","G","G06","G06F","G06F16/00","G06F16/20","G06F16/27","G","G06","G06F","G06F16/00","G06F16/20","G06F16/25","G06F16/254","G","G06","G06F","G06F21/00","G06F21/60","G06F21/602","G","G06","G06N","G06N20/00","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G","G06","G06Q","G06Q10/00","G06Q10/08","G","G06","G06Q","G06Q30/00","G06Q30/01","G","G06","G06Q","G06Q30/00","G06Q30/018","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0207","G","G06","G06Q","G06Q30/00","G06Q30/06","G","G06","G06Q","G06Q2220/00"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201724A1/en"},{"publication_number":"AU2026201642A1","title":"Systems and methods for auditing assets","abstract":"In one embodiment, a method includes receiving first Light Detection and Ranging (LiDAR) data associated with a railroad environment, extracting an asset from the first LiDAR data associated with the railroad environment, and superimposing the asset into a spatial model. The method also includes receiving a field indication associated with a modification to the railroad environment and modifying the spatial model in response to receiving the field indication associated with the modification to the railroad environment. The method further includes receiving second LiDAR data associated with the railroad environment and comparing the second LiDAR data to the modified spatial model. 20 26 20 16 42 04 M ar 2 02 6 2 0 2 6 2 0 1 6 4 2 0 4 M a r 2 0 2 6","assignee":"BNSF Railway Co","inventors":["Philip Gerald Davis","Jacob Scott Ford","Ross Aaron Springer","Devin Scott Wagner"],"publication_date":"2026-03-26","filing_date":"2026-03-04","priority_date":"2019-10-16","cpc_codes":["B","B61","B61K","B61K9/00","B61K9/08","B","B61","B61L","B61L23/00","B61L23/04","B61L23/041","B","B61","B61L","B61L23/00","B61L23/04","B61L23/042","B","B61","B61L","B61L25/00","B61L25/02","B61L25/025","B","B61","B61L","B61L25/00","B61L25/06","G","G01","G01S","G01S17/00","G01S17/88","G01S17/89","G","G01","G01S","G01S7/00","G01S7/48","G01S7/4802","G","G01","G01S","G01S7/00","G01S7/48","G01S7/4808","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q50/00","G06Q50/40"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201642A1/en"},{"publication_number":"AU2026201656A1","title":"Plane wave dual basis for quantum simulation","abstract":"22 Methods, systems and apparatus for simulating quantum systems. In one aspect, a method includes the actions of obtaining a first Hamiltonian describing the quantum system, wherein the Hamiltonian is written in a plane wave basis comprising N plane wave basis vectors; applying a discrete Fourier transform to the first Hamiltonian to generate a second Hamiltonian written in a plane wave dual basis, wherein the second Hamiltonian comprises a number of terms that scales at most quadratically with N; and simulating the quantum system using the second Hamiltonian. 61687736.docx","assignee":"Google LLC","inventors":["Ryan BABBUSH"],"publication_date":"2026-03-26","filing_date":"2026-03-04","priority_date":"2017-05-19","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/20","G","G06","G06F","G06F17/00","G06F17/10","G06F17/14","G06F17/141","G","G06","G06E","G06E3/00","G06E3/001","G06E3/005","G","G06","G06F","G06F17/00","G06F17/10","G06F17/18","G","G06","G06N","G06N10/00","G","G06","G06N","G06N10/00","G06N10/20","G","G06","G06N","G06N10/00","G06N10/60","H","H04","H04L","H04L9/00","H04L9/08","H04L9/0816","H04L9/0852","G","G06","G06F","G06F2111/00","G06F2111/10"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201656A1/en"},{"publication_number":"AU2026201660A1","title":"Using a set of machine learning diagnostic models to determine a diagnosis based on a skin tone of a patient","abstract":"Systems and methods are disclosed herein for determining a diagnosis based on a base skin tone of a patient. In an embodiment, the system receives a base skin tone image of a patient, generates a calibrated base skin tone image by calibrating the base skin tone image using a reference calibration profile, and determines a base skin tone of the patient based on the calibrated base skin tone image. The system receives a concern image of a portion of the patient’s skin, and selects a set of machine learning diagnostic models from a plurality of sets of candidate machine learning diagnostic models based on the base skin tone of the patient, each of the sets of candidate machine learning diagnostic models trained to receive the concern image and output a diagnosis of a condition of the patient.","assignee":"Digital Diagnostics Inc","inventors":["Elektra Efstratiou ALIVISATOS","Elizabeth Asai","Joseph Ferrante","Elliot Swart"],"publication_date":"2026-03-26","filing_date":"2026-03-04","priority_date":"2019-06-18","cpc_codes":["A","A61","A61B","A61B5/00","A61B5/0002","A61B5/0004","A61B5/0013","A","A61","A61B","A61B5/00","A61B5/0059","A61B5/0077","A","A61","A61B","A61B5/00","A61B5/103","A61B5/1032","A","A61","A61B","A61B5/00","A61B5/44","A61B5/441","A","A61","A61B","A61B5/00","A61B5/44","A61B5/441","A61B5/444","A","A61","A61B","A61B5/00","A61B5/68","A61B5/6887","A61B5/6898","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7275","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/24765","G","G06","G06N","G06N20/00","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G06T7/0014","G","G06","G06T","G06T7/00","G06T7/90","G","G06","G06V","G06V10/00","G06V10/40","G06V10/56","G","G06","G06V","G06V10/00","G06V10/70","G06V10/72","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V40/00","G06V40/10","G","G16","G16H","G16H30/00","G16H30/40","G","G16","G16H","G16H40/00","G16H40/60","G16H40/67","G","G16","G16H","G16H50/00","G16H50/20","A","A61","A61B","A61B2576/00","A61B2576/02","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10024","G"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201660A1/en"},{"publication_number":"KR20260040208A","title":"System for predicting ocean and climate environmental data based on time-series data reflecting interactions among environmental variables","abstract":"전자 장치의 동작 방법이 개시된다. 본 개시에 따른 전자 장치의 동작 방법은, 일 해양 구역에 대한 복수의 환경 변수의 값을 실시간으로 획득하는 단계 및 다중 LSTM(Long-Short Term Memory) 구조 상에 실시간으로 획득된 복수의 환경 변수의 값을 입력하여, 시계열에 따른 복수의 환경 변수 각각의 값을 예측하는 단계를 포함한다. A method of operating an electronic device is disclosed. The method of operating an electronic device according to the present disclosure includes the step of acquiring values of a plurality of environmental variables for a marine area in real time, and the step of inputting the values of the plurality of environmental variables acquired in real time onto a multiple LSTM (Long-Short Term Memory) structure to predict the value of each of the plurality of environmental variables according to a time series.","assignee":"주식회사 시즈","inventors":["이미애","이호근","함동빈"],"publication_date":"2026-03-24","filing_date":"2026-03-06","priority_date":"2024-09-06","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G01","G01C","G01C13/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260040208A/en"},{"publication_number":"KR20260040203A","title":"PPG-Based 12-Lead ECG Reconstruction System and Method","abstract":"본 발명은 광용적맥파(PPG) 신호로 임상 표준 12-Lead 심전도를 높은 정확도로 재구성할 수 있는 시스템에 관한 것이다. 본 시스템은 PPG 센서를 통해 PPG 신호를 취득하여 대역통과 필터링 및 정규화를 수행한다. 전처리된 PPG 신호를 딥러닝 모델을 이용하여 Lead I 및 Lead II 심전도 신호로 변환하고 생성된 Lead I 및 Lead II 심전도 신호로부터 나머지 10개의 심전도 신호를 재구성한다. 이 때, 사지유도는 보정된 아인트호벤 알고리즘을 적용하고, 흉부유도는 2채널 입력 다중 시드 앙상블 및 확률적 가중치 평균(SWA)을 이용하여 6개의 흉부유도 심전도 신호를 생성한다. The present invention relates to a system capable of reconstructing a clinical standard 12-lead electrocardiogram with high accuracy using photoplethysmography (PPG) signals. The system acquires PPG signals through a PPG sensor and performs bandpass filtering and normalization. The preprocessed PPG signals are converted into Lead I and Lead II electrocardiogram signals using a deep learning model, and the remaining 10 electrocardiogram signals are reconstructed from the generated Lead I and Lead II electrocardiogram signals. At this time, the limb leads are subjected to a corrected Eindhoven algorithm, and the chest leads are subjected to a 2-channel input multi-seed ensemble and stochastic weighted average (SWA) to generate 6 chest lead electrocardiogram signals.","assignee":"주식회사 자하랩스","inventors":["한병희"],"publication_date":"2026-03-24","filing_date":"2026-03-05","priority_date":"2026-03-05","cpc_codes":["A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7278","A","A61","A61B","A61B5/00","A61B5/02","A61B5/024","A61B5/02416","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7225","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G16","G16H","G16H50/00","G16H50/20"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260040203A/en"},{"publication_number":"KR20260040401A","title":"Ai-based patent infringement judgment and automated response system and method thereof","abstract":"본 발명은 AI 기반 특허 침해 판단 및 자동 대응 시스템에 관한 것으로, 대상 기술 문헌 및 기 등록된 특허 문헌으로부터 기술적 특징점을 추출하는 데이터 분석부, 상기 추출된 특징점을 비교 분석하여 침해 확률 정보 및 침해 증거 데이터를 생성하는 침해 판단부, 상기 침해 판단부에서 생성된 침해 확률 정보 및 침해 증거 데이터를 블록체인 네트워크 외부에서 내부로 전달하되, 복수의 노드로부터 데이터의 무결성이 검증된 값을 전달하는 오라클 인터페이스부, 및 상기 오라클 인터페이스부를 통해 수신된 침해 확률 정보가 미리 설정된 임계값을 초과하는 경우, 해당 기술권자에게 알림을 발송하거나 해당 대상 기술의 사용 제한 또는 라이선스 계약을 자동 집행하는 스마트 컨트랙트부를 포함한다. The present invention relates to an AI-based patent infringement determination and automatic response system, comprising: a data analysis unit that extracts technical feature points from target technical literature and previously registered patent literature; an infringement determination unit that generates infringement probability information and infringement evidence data by comparing and analyzing the extracted feature points; an oracle interface unit that transmits the infringement probability information and infringement evidence data generated by the infringement determination unit from outside to inside a blockchain network, wherein the oracle interface unit transmits values in which the integrity of the data has been verified by a plurality of nodes; and a smart contract unit that, when the infringement probability information received through the oracle interface unit exceeds a preset threshold, sends a notification to the relevant technology right holder or automatically executes a restriction on the use of the target technology or a license agreement.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-03-24","filing_date":"2026-02-28","priority_date":"2026-02-28","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/18","G06Q50/184","G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06F","G06F40/00","G06F40/40","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0207","G","G08","G08B","G08B25/00","H","H04","H04L","H04L9/00","H04L9/50"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260040401A/en"},{"publication_number":"KR20260040207A","title":"Apparatus and system for efficient prediction of real-time input data based on interoperation between a multi-lstm model and a transformer model","abstract":"전자 장치의 동작 방법이 개시된다. 본 개시에 따른 전자 장치의 동작 방법은, 일 해양 구역에 대한 복수의 환경 변수의 값을 실시간으로 획득하는 단계 및 다중 LSTM(Long-Short Term Memory) 구조 상에 실시간으로 획득된 복수의 환경 변수의 값을 입력하여, 시계열에 따른 복수의 환경 변수 각각의 값을 예측하는 단계를 포함한다. A method of operating an electronic device is disclosed. The method of operating an electronic device according to the present disclosure includes the step of acquiring values of a plurality of environmental variables for a marine area in real time, and the step of inputting the values of the plurality of environmental variables acquired in real time onto a multiple LSTM (Long-Short Term Memory) structure to predict the value of each of the plurality of environmental variables according to a time series.","assignee":"주식회사 시즈","inventors":["이미애","이호근","함동빈"],"publication_date":"2026-03-24","filing_date":"2026-03-06","priority_date":"2024-09-06","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260040207A/en"},{"publication_number":"CN121559504B","title":"A method and system for synthesizing reflectivity of weather radar based on geostationary satellites","abstract":"本发明涉及天气雷达合成技术领域，尤其是涉及一种基于同步静止卫星的天气雷达反射率合成方法及系统。方法包括获取卫星多通道观测数据和天气雷达数据；基于获取的卫星多通道观测数据和天气雷达数据进行数据预处理；构建基于改进Swin‑Transformer和U‑Net的天气雷达合成模型；对构建的天气雷达合成模型进行模型训练；利用训练后的模型进行前向推理并得到天气雷达反射率合成。本发明提供了对这些区域的高质量天气观测数据，解决了天气雷达数据空缺的问题，显著提升了天气雷达观测范围，尤其是对于大尺度天气系统和强对流天气的监测能力。 This invention relates to the field of weather radar synthesis technology, and in particular to a method and system for synthesizing weather radar reflectivity based on geostationary satellites. The method includes acquiring multi-channel satellite observation data and weather radar data; performing data preprocessing based on the acquired multi-channel satellite observation data and weather radar data; constructing a weather radar synthesis model based on an improved Swin-Transformer and U-Net; training the constructed weather radar synthesis model; and using the trained model for forward inference to obtain the synthesized weather radar reflectivity. This invention provides high-quality weather observation data for these regions, solves the problem of missing weather radar data, and significantly improves the observation range of weather radar, especially its ability to monitor large-scale weather systems and severe convective weather.","assignee":"Ocean University of China","inventors":["张巍","马宏博","王贤俊"],"publication_date":"2026-03-24","filing_date":"2026-01-22","priority_date":"2026-01-22","cpc_codes":["G","G01","G01S","G01S13/00","G01S13/86","G","G01","G01S","G01S13/00","G01S13/88","G01S13/95","G","G01","G01S","G01S7/00","G01S7/02","G01S7/41","G","G01","G01W","G01W1/00","G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N5/00","G06N5/04","G06N5/046","Y","Y02","Y02A","Y02A90/00","Y02A90/10"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121559504B/en"},{"publication_number":"CN121562678B","title":"A Heterogeneous Hardware Adaptive Acceleration Method and System Based on Reinforcement Learning for SM4","abstract":"The invention discloses a heterogeneous hardware SM4 self-adaptive acceleration method based on reinforcement learning, which models the SM4 self-adaptive acceleration problem in a heterogeneous environment as a reinforcement learning process, utilizes a strong nonlinear fitting capability of a deep network and a near-end strategy optimization algorithm to find an optimal solution in a complex software and hardware combination space through a closed loop link of full stack perception-intelligent decision-dynamic library call-feedback evolution, and simultaneously shields the difference of a bottom heterogeneous instruction set through a preset binary library so that an upper-layer intelligent agent can concentrate on strategy optimization, thereby realizing the unification of performance and flexibility. The method can solve the technical problem that the complexity of the heterogeneous Internet of things environment cannot be effectively treated by the existing specific instruction set acceleration method based on static compiling optimization, and the technical problem that the existing running time-division method based on simple heuristic rules lacks the sensing and adaptation capability to the dynamic load of the system.","assignee":"Hunan Kuangan Network Technology Co ltd","inventors":["吕婷","肖紫东","曾旺","蔡宇辉","杨志邦","余思洋","杨圣洪","李肯立","唐伟","段明星","王卫民"],"publication_date":"2026-03-24","filing_date":"2026-01-22","priority_date":"2026-01-22","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","H","H04","H04L","H04L63/00","H04L63/04","H04L63/0428","H","H04","H04L","H04L9/00","H04L9/06","H04L9/0643","H","H04","H04L","H04L9/00","H04L9/40","H","H04","H04L","H04L2209/00","H04L2209/12","H04L2209/122"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121562678B/en"},{"publication_number":"CN121561164B","title":"Multi-mode alignment method based on text prior and asymmetric depth mixing","abstract":"The invention belongs to the technical field of multi-modal fusion alignment, and discloses a multi-modal alignment method based on text priori and asymmetric depth mixing. The method comprises a mixed query construction module, a Cat-MoD model and a similarity perception routing supervision module, wherein the mixed query construction module fuses a guide token with text priori and an exploration token, the Cat-MoD model comprises a plurality of asymmetric depth mixing modules, an instruction sequence and a token sequence processed by the Cat-MoD model are input into a frozen large model together to be processed to obtain a final output result, the similarity perception routing supervision module calculates cosine similarity of the query token in the asymmetric depth mixing modules before and after interaction with picture features and uses the cosine similarity as a label to provide supervision signals for routers in the asymmetric depth mixing modules, and the asymmetric depth mixing modules evaluate information saturation of the query token input into the current asymmetric depth mixing modules through the routers.","assignee":"东北大学","inventors":["黄奕杰","冯时","王大玲","张一飞","杨晓翠"],"publication_date":"2026-03-24","filing_date":"2026-01-21","priority_date":"2026-01-21","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/90335","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9032","G06F16/90332","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121561164B/en"},{"publication_number":"CN121542657B","title":"Low-overhead sampling analysis method for broadband wireless signals","abstract":"The invention relates to a low-overhead sampling analysis method for broadband wireless signals, which comprises the following steps of obtaining an original low-rate IQ sample sequence of a target broadband signal, carrying out multidimensional feature embedding to obtain an original feature sequence, carrying out autocorrelation analog sampling treatment on the original low-rate IQ sample sequence to obtain an enhanced autocorrelation feature sequence, splicing the enhanced autocorrelation feature sequence with the original feature sequence to form a fused enhancement feature sequence, inputting the fused enhancement feature sequence into a spectrum sensing model based on a deep neural network for processing, predicting spectrum occupation priori information in real time, reconstructing the original low-rate IQ sample sequence as a signal reconstruction constraint to obtain an effective recovery signal, and carrying out signal analysis on the effective recovery signal by adopting a signal analysis model based on a transducer to obtain a deep analysis result. Compared with the prior art, the invention has the advantages of breaking through the limitation of the sampling rate in the prior art, greatly simplifying the data processing flow and the like.","assignee":"Fudan University","inventors":["高跃","陈哲","彭劲搏"],"publication_date":"2026-03-24","filing_date":"2026-01-21","priority_date":"2026-01-21","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121542657B/en"},{"publication_number":"CN121542651B","title":"Vehicle energy consumption prediction method, device and medium based on double-branch input","abstract":"The application discloses a vehicle energy consumption prediction method, device and medium based on double-branch input, relating to the technical field of vehicle energy consumption prediction, wherein the method comprises the steps of carrying out multi-mode feature extraction on driving energy consumption data to obtain a normalized time sequence; the method comprises the steps of carrying out multistage discrete wavelet transformation on a normalized time sequence to determine a standard multichannel input tensor, extracting local features of the standard multichannel input tensor to obtain a local space-time feature sequence, calculating and splicing hidden states of the local space-time feature sequence on each time step to determine a hidden state sequence, calculating attention weights corresponding to the hidden state sequence to obtain a context vector, determining an energy consumption feature vector according to a macroscopic working condition feature vector and the context vector, and carrying out feature aggregation and energy consumption prediction mapping on the energy consumption feature vector to obtain an energy consumption predicted value of a target vehicle. The method solves the technical problems of insufficient robustness and accuracy of vehicle energy consumption prediction in the prior art.","assignee":"Qingdao Qingte Zhongli Axle Co ltd; Shandong University of Science and Technology","inventors":["孟凡钰","纪玉龙","王健","扈建龙","赵子亮","任传祥"],"publication_date":"2026-03-24","filing_date":"2026-01-19","priority_date":"2026-01-19","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2131","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q50/00","G06Q50/06","G","G06","G06Q","G06Q50/00","G06Q50/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121542651B/en"},{"publication_number":"CN121526883B","title":"Hyperspectral image super-resolution method and device based on iterative reconstruction and spectrum refining","abstract":"The invention provides a hyperspectral image super-resolution method and a hyperspectral image super-resolution device based on iterative reconstruction and spectrum refining, which relate to the technical field of hyperspectral images, wherein a three-adjacent window cube is firstly intercepted in a sliding mode along a spectrum dimension and mapped into a pseudo RGB three-channel feature through a lightweight head adapter; the method comprises the steps of performing domain adaptation on a frozen pre-training network through LoRA low-rank decomposition, outputting corresponding three-band high-resolution prediction, performing normalization fusion on multiple predictions generated by different center windows of the same target band through spectrum position window weight multiplied by confidence level to obtain a coarse superresolution result, and performing end-to-end refinement under the constraint of a combined SAM spectrum angle by a spectrum reconstruction decoder to obtain a final high-resolution hyperspectral image. The method aims to realize the synchronous jump of space details and spectrum fidelity by carrying out lossless migration on a natural image pre-training model to a hundred-band hyperspectral scene with extremely low parameter increment.","assignee":"Xiamen University of Technology","inventors":["杜晓凤","颜龙","曹卫","林仙丽"],"publication_date":"2026-03-24","filing_date":"2026-01-16","priority_date":"2026-01-16","cpc_codes":["G","G06","G06T","G06T3/00","G06T3/40","G06T3/4053","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06T","G06T3/00","G06T3/40","G06T3/4046","G","G06","G06T","G06T3/00","G06T3/40","G06T3/4053","G06T3/4076"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121526883B/en"},{"publication_number":"CN121501827B","title":"A multimodal knowledge graph-based question-and-answer method for pests and diseases that integrates agricultural environmental characteristics","abstract":"本发明公开了一种融合农业环境特征的多模态知识图谱病虫害问答方法，属于人工智能与智慧农业技术领域，包括：收集文本和图像，对文本和图像进行预处理，将预处理后的文本与图像利用蒸馏MiniCPM‑V 4.5模型得到特征文本、特征图像和视觉识别结果，再利用农业知识图谱进行检索，得到实体、实体所在社区、农业知识导向摘要和防治核心措施，再结合视觉识别结果得到结构化答案，再利用端侧界面进行展示得到精准咨询需求答案展示。本发明解决现有系统数据专业性不足、基于图谱的检索增强生成领域适配差、知识抽取精度低的问题，微调后病虫害识别准确率达93.3%，可部署于端侧设备，为基层从业者提供高效咨询服务。 This invention discloses a multimodal knowledge graph-based question-and-answer method for pests and diseases that integrates agricultural environmental characteristics. Belonging to the fields of artificial intelligence and smart agriculture, the method includes: collecting text and images; preprocessing the text and images; using the MiniCPM-V 4.5 model to obtain feature text, feature images, and visual recognition results; then using an agricultural knowledge graph for retrieval to obtain entities, their communities, agricultural knowledge-guided summaries, and core prevention and control measures; finally, combining the visual recognition results to obtain structured answers; and finally, displaying the answers to precise consultation needs on a mobile interface. This invention addresses the problems of insufficient data professionalism, poor domain adaptation of graph-based retrieval enhancement generation, and low knowledge extraction accuracy in existing systems. After fine-tuning, the accuracy rate of pest and disease identification reaches 93.3%, and it can be deployed on mobile devices to provide efficient consultation services for grassroots practitioners.","assignee":"Wuhan University of Technology WUT","inventors":["熊盛武","胡志康","段鹏飞"],"publication_date":"2026-03-24","filing_date":"2026-01-14","priority_date":"2026-01-14","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/242","G06F16/243","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2455","G06F16/24564","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/248","G","G06","G06F","G06F16/00","G06F16/20","G06F16/25","G06F16/254","G","G06","G06F","G06F16/00","G06F16/50","G06F16/58","G06F16/583","G","G06","G06F","G06F40/00","G06F40/30","G06F40/35","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06Q","G06Q50/00","G06Q50/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121501827B/en"},{"publication_number":"CN121503621B","title":"A method and apparatus for revising knowledge ontology beliefs based on a reinforcement learning-based agent model.","abstract":"本发明涉及人工智能技术领域，公开了一种基于强化学习的智能体模型的知识本体信念修正方法及装置，用于解决现有技术对智能体模型的知识本体进行信念修正的方式不够智能、合理，导致修正后的知识体系易出现矛盾或者偏离事实情况的技术问题。该方法包括：获取智能体模型运行时中发现的新增知识断言；检测新增知识断言是否与智能体模型的知识本体中原有知识断言存在冲突；若是，则基于新增知识断言和存在冲突的原有知识断言生成最小冲突断言集；基于最小冲突断言集求解出修正策略集；调用基于强化学习算法构建的策略选择模型，在修正策略集中选择目标修正策略；基于目标修正策略对知识本体进行修正，并将信念修正后的知识本体注入智能体模型。 This invention relates to the field of artificial intelligence technology and discloses a method and apparatus for correcting the belief of a knowledge ontology in an intelligent agent model based on reinforcement learning. This method addresses the technical problem that existing methods for correcting the belief of a knowledge ontology in an intelligent agent model are not intelligent or reasonable enough, leading to contradictions or deviations from reality in the corrected knowledge system. The method includes: acquiring newly added knowledge assertions discovered during the operation of the intelligent agent model; detecting whether the newly added knowledge assertions conflict with existing knowledge assertions in the knowledge ontology of the intelligent agent model; if so, generating a minimum conflict assertion set based on the newly added knowledge assertions and the conflicting existing knowledge assertions; solving for a correction strategy set based on the minimum conflict assertion set; calling a strategy selection model built based on a reinforcement learning algorithm to select a target correction strategy from the correction strategy set; correcting the knowledge ontology based on the target correction strategy; and injecting the belief-corrected knowledge ontology into the intelligent agent model.","assignee":"Zhiwei Xingyi Shanghai Intelligent Technology Co ltd","inventors":["程煜轩","施磊"],"publication_date":"2026-03-24","filing_date":"2026-01-13","priority_date":"2026-01-13","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121503621B/en"},{"publication_number":"CN121505461B","title":"Support vector machine-based rice seed classification method and system","abstract":"A method and a system for classifying rice seeds based on a support vector machine belong to the technical field of computation intelligence and optimization algorithms, and solve the technical problem that the accuracy of classifying rice seeds is reduced because the optimal solution to the problem is missed due to the fact that the parameters of the support vector machine are selected according to experience in the prior art. Collecting hyperspectral image data of the rice japonica seeds, initializing a rime optimizing algorithm based on a fused elite optimizing strategy to generate an initial solution, adopting a hidden mechanism to improve a greedy strategy to update a population, optimizing punishment coefficients and kernel function parameters in a support vector machine by using the improved rime optimizing algorithm, designing a mapping rule to carry out iterative optimization on the kernel functions in the support vector machine to obtain an optimal solution, setting the optimal solution as the parameters of the support vector machine, and constructing a rice japonica seed variety classification model to realize rice japonica seed variety classification. The invention is used for realizing high-efficiency and accurate classification of the rice seeds and the japonica rice seeds.","assignee":"Jilin Agricultural University","inventors":["于合龙","张嘉豪","陈振洋","张康","何蕊","周雷进雨"],"publication_date":"2026-03-24","filing_date":"2026-01-13","priority_date":"2026-01-13","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/10","G06V20/194","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121505461B/en"},{"publication_number":"KR20260039961A","title":"Apparatus and method for generating a depth map using a volumetric feature","abstract":"본 개시는 볼류메트릭 피처를 이용하여 깊이 맵을 생성하는 장치 및 방법에 관한 것이다. 본 개시의 일 실시 예에 따른 방법은, 서라운드 뷰 이미지에 포함된 기본 이미지를 인코딩 및 후처리하여 기본 이미지에 대한 단일 피처맵을 생성하고, 단일 피처맵을 깊이 정보와 함께 인코딩한 후, 인코딩 결과를 3차원 공간으로 투영함으로써 볼류메트릭 피처를 생성할 수 있다. 본 방법에서는 깊이 디코더를 이용하여 볼류메트릭 피처를 디코딩함으로써, 서라운드 뷰 이미지의 깊이 맵을 생성할 수 있다. The present disclosure relates to an apparatus and method for generating a depth map using volumetric features. A method according to one embodiment of the present disclosure can generate a volumetric feature by encoding and post-processing a base image included in a surround view image to generate a single feature map for the base image, encoding the single feature map together with depth information, and then projecting the encoding result into a three-dimensional space. In the present method, a depth map of a surround view image can be generated by decoding the volumetric feature using a depth decoder.","assignee":"포티투닷 주식회사","inventors":["김중희","푸옥 응우옌 티엔","정성균","허준화"],"publication_date":"2026-03-23","filing_date":"2026-03-04","priority_date":"2022-07-29","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/50","G06T7/55","G06T7/593","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T3/00","G06T3/40","G","G06","G06T","G06T7/00","G06T7/70","H","H04","H04N","H04N13/00","H04N13/10","H04N13/106","H04N13/161","H","H04","H04N","H04N13/00","H04N13/20","H04N13/204","H04N13/207","H","H04","H04N","H04N13/00","H04N13/30","H04N13/388","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30244"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260039961A/en"},{"publication_number":"KR20260039968A","title":"Electron Trap Device for Quantum Computers","abstract":"본 발명은 양자 컴퓨터용 전자 트랩 장치로서, 기판; 상기 기판 상에 형성된 액체 헬륨 막; 상기 액체 헬륨 막의 상면(upper surface) 상에 배치되며, 적어도 하나의 전자를 포획하여 큐비트로 동작시키도록 구성된 전자 트랩 영역; 상기 액체 헬륨 막의 하면(lower surface) 아래에 배치되며, 상기 전자 트랩 영역으로 공급될 복수의 전자를 유지하는 전자 저장소; 및 상기 전자 저장소로부터 상기 전자 트랩 영역으로 전자를 수송하기 위해 상기 액체 헬륨 막을 수직 방향으로 관통하여 형성된 수직 전자 이동 채널을 포함하며, 상기 전자 저장소는 상기 액체 헬륨 막에 의해 상기 전자 트랩 영역과 수직으로 공간 격리됨으로써, 상기 전자 트랩 영역에 가해지는 마이크로파(microwave) 신호가 상기 전자 저장소 내의전자에 미치는 간섭을 차단하는 것을 특징으로 하는, 양자 컴퓨터용 전자 포획 장치가 제공된다. The present invention provides an electron trap device for a quantum computer, comprising: a substrate; a liquid helium film formed on the substrate; an electron trap region disposed on the upper surface of the liquid helium film and configured to capture at least one electron and operate it as a qubit; an electron storage unit disposed below the lower surface of the liquid helium film and holding a plurality of electrons to be supplied to the electron trap region; and a vertical electron transport channel formed by penetrating the liquid helium film in a vertical direction to transport electrons from the electron storage unit to the electron trap region, wherein the electron storage unit is spatially isolated vertically from the electron trap region by the liquid helium film, thereby blocking interference between a microwave signal applied to the electron trap region and an electron within the electron storage unit.","assignee":"안범주","inventors":["안범주"],"publication_date":"2026-03-23","filing_date":"2026-03-04","priority_date":"2026-03-04","cpc_codes":["H","H10","H10D","H10D48/00","H10D48/383","G","G06","G06N","G06N10/00","G06N10/40","H","H10","H10D","H10D62/00","H10D62/80","H10D62/81","H","H10","H10D","H10D64/00","H10D64/01","H10D64/031","H10D64/037","H","H10","H10N","H10N70/00","H10N70/20","H10N70/25","H","H10","H10N","H10N99/00","H10N99/05"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260039968A/en"},{"publication_number":"KR20260039951A","title":"Operational Standard Integration via Quantified Conflict Control","abstract":"본 발명은 복수의 운영 기준이 상충되는 환경에서, 기준 간 충돌을 정량화하고 이를 자동으로 조정하여 단일 통합 운영 기준을 생성하는 시스템에 관한 것이다. 본 발명에 따르면, 기준 충돌로 인한 판단 불일치를 최소화하고 운영 상황에 적합한 일관된 기준을 자동으로 제공함으로써, 운영 안정성 및 관리 신뢰성을 효과적으로 향상시킬 수 있다. The present invention relates to an environment where multiple operating standards conflict, Quantify conflicts between standards and automatically adjust them It is about a system that creates a single integrated operational standard. According to the present invention, judgment discrepancies caused by reference conflicts are minimized. By automatically providing consistent standards suitable for the operating situation, It can effectively improve operational stability and management reliability.","assignee":"구교선; 구현우","inventors":["구교선","구현우"],"publication_date":"2026-03-23","filing_date":"2026-02-23","priority_date":"2026-02-23","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06316","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0633","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06395","G","G08","G08B","G08B21/00","G08B21/18"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260039951A/en"},{"publication_number":"KR20260039950A","title":"System and Method for Diagnosing Osteoporosis Using X-ray Images","abstract":"일실시예에 따른 골다공증 진단 장치는. 전체 영상을 입력 받아 제1 로짓을 출력하는 제1 모델; 상기 전체 영상을 해부학적 부위에 대응하여 분할한 복수의 분할영상을 각각 입력 받아 복수의 제n 로짓을 각각 출력하는 복수의 제n 모델(n은 2이상의 정수); 및 상기 제1 로짓과 상기 복수의 제n 로짓에 기초하여 골다공증 여부를 식별하는 골 질환 분류 추론부를 포함한다. 상기 전체 영상은 정상 레이블 또는 골다공증 레이블 또는 골감소증 레이블 중 적어도 2 개 이상으로 구분되는 레이블 중 어느 하나로 매칭되고, 상기 복수의 분할영상은 상기 전체 영상에 매칭된 레이블과 동일하게 매칭될 수 있다. An osteoporosis diagnostic device according to one embodiment comprises: a first model that receives an entire image and outputs a first logit; a plurality of n-th models (where n is an integer greater than or equal to 2) that each receive a plurality of segmented images corresponding to anatomical regions and each output a plurality of n-th logits; and a bone disease classification inference unit that identifies whether or not there is osteoporosis based on the first logit and the plurality of n-th logits. The entire image above is matched to any one of at least two labels, such as a normal label, an osteoporosis label, or an osteopenia label, and the plurality of segmented images can be matched to the same label as the label matched to the entire image.","assignee":"프로메디우스 주식회사","inventors":["장미소","이가은"],"publication_date":"2026-03-23","filing_date":"2026-02-13","priority_date":"2024-04-19","cpc_codes":["A","A61","A61B","A61B6/00","A61B6/52","A61B6/5211","A61B6/5217","A","A61","A61B","A61B6/00","A61B6/50","A61B6/505","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G","G06","G06T","G06T7/00","G06T7/10","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G16","G16H","G16H30/00","G16H30/40","G","G16","G16H","G16H50/00","G16H50/20","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10116","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30004","G06T2207/30008"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260039950A/en"},{"publication_number":"TR2026001234A2","title":"THREE-DIMENSIONAL PARAMETRIC KÜNDEKÂRÎ PATTERN PRODUCTION SYSTEM AND METHOD","abstract":"Buluş, kündekârî desenlerinin kural tabanlı, algoritmik ve parametrik yöntemlerle üç boyutlu olarak üretilmesini ve modellenmesini sağlayan dijital bir sistem ve yöntemdir. Daha spesifik olarak buluş; kündekârî desenlerinin algoritmik yöntemlerle analiz edilmesini, biçim grameri kuralları ile tanımlanmasını ve parametrik modelleme ortamında yeniden üretilmesini sağlayan özgün bir yöntemdir. The invention is a digital system and method that enables the three-dimensional generation and modeling of kündekârî patterns using rule-based, algorithmic, and parametric methods. More specifically, the invention is a unique method that allows for the algorithmic analysis of kündekârî patterns, their definition using form grammar rules, and their reproduction in a parametric modeling environment.","assignee":"Van Yüzüncü Yil Üni̇versi̇tesi̇","inventors":["Bayer Semi̇h"],"publication_date":"2026-03-23","filing_date":"2026-01-28","priority_date":"2026-01-28","cpc_codes":["G","G06","G06N","G06N5/00","G","G06","G06T","G06T17/00"],"country":"TR","kind":"application","source_url":"https://patents.google.com/patent/TR2026001234A2/en"},{"publication_number":"TR2026000957A2","title":"A REDOX AND ANTIOXIDANT CAPACITY MONITORING SYSTEM AND METHOD OF ACTIVITY","abstract":"Buluş, redoks biyolojisi, toksikoloji, ilaç geliştirme ve hücresel stres araştırmaları alanlarında kullanılmak üzere tasarlanmış, fizyolojik oksijen düzeylerinde (1?8 kPa O?) çalışan, hayvan kullanılmadan hücresel redoks yanıtı ve antioksidan kapasiteyi nicel olarak ölçen yüksek-içerikli in-vitro platformu olan bir redoks ve antioksidan kapasitesi takip sistemi ve çalışma yöntemi ile ilgilidir. The invention relates to a redox and antioxidant capacity monitoring system and study method, a high-content in-vitro platform designed for use in redox biology, toxicology, drug development, and cellular stress research, operating at physiological oxygen levels (1–8 kPa O₂) and quantitatively measuring cellular redox response and antioxidant capacity without the use of animals.","assignee":"İstanbul Medi̇pol Üni̇versi̇tesi̇","inventors":["Eroğlu Emrah","Kati Ahmet","Armouch Joudi","Kök Kivanç","Issa Hamzah","Yildirim Sena"],"publication_date":"2026-03-23","filing_date":"2026-01-22","priority_date":"2026-01-22","cpc_codes":["C","C12","C12Q","C12Q1/00","C12Q1/02","C","C12","C12M","C12M1/00","C12M1/34","C","C12","C12M","C12M3/00","C","C12","C12N","C12N5/00","C12N5/0018","G","G01","G01N","G01N21/00","G01N21/62","G01N21/63","G01N21/64","G","G01","G01N","G01N33/00","G01N33/48","G01N33/50","G01N33/5005","G01N33/5008","G01N33/5044","G01N33/5047","G","G06","G06N","G06N20/00"],"country":"TR","kind":"application","source_url":"https://patents.google.com/patent/TR2026000957A2/en"},{"publication_number":"TR2025022468A2","title":"EDGE AI AND DISTRIBUTED COMPUTING - DYNAMIC MODEL SPLITTING AND LOW-LAT INFERENCE SYSTEM WITH SPLIT LEARNING FRAMEWORK","abstract":"Bu buluş, edge AI tabanlı dağıtık çıkarımda split learning ile dinamik model bölümleme alanına ilişkindir. Çözülen problem, statik partition ve basit offloading kararlarının değişken ağ/cihaz koşullarında gecikme artışı ve gizlilik maliyeti üretmesidir. Buluş, katman maliyeti profilini çıkarıp öğrenme tabanlı partition kararı vererek modeli cihaz ve MEC/bulut arasında istek bazında böler. Sıkıştırma, gizlilik şifreleme ve erken çıkış mekanizmalarıyla bant genişliği, gecikme ve gizlilik hedeflerini birlikte optimize eder. This invention relates to dynamic model partitioning with split learning in edge AI-based distributed inference. The problem solved is that static partitioning and simple offloading decisions produce increased latency and privacy costs under variable network/device conditions. The invention partitions the model on a request basis between the device and the MEC/cloud by extracting the layer cost profile and making learning-based partitioning decisions. It optimizes bandwidth, latency, and privacy objectives simultaneously through compression, privacy encryption, and early exit mechanisms.","assignee":"Türk Telekomüni̇kasyon Anoni̇m Şi̇rketi̇","inventors":["Mustafa Aydin Hürkan"],"publication_date":"2026-03-23","filing_date":"2025-12-29","priority_date":"2025-12-29","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"TR","kind":"application","source_url":"https://patents.google.com/patent/TR2025022468A2/en"},{"publication_number":"TR2025021656A1","title":"AI-SUPPORTED AUTONOMOUS ROOT CAUSE ANALYSIS AND ANOMALY REMOVAL METHOD IN TELECOMMUNICATIONS MEDIATION PROCESSES AND A RELATED SYSTEM","abstract":"Buluş, telekomünikasyon arabuluculuk (mediation) süreçlerinde meydana gelen karmaşık veri anomalilerini ve işlem hatalarını (örneğin, reddedilen CDR kayıtlarını), manuel müdahaleye gerek kalmadan yönetebilmeyi sağlayan yapay zekâ destekli otonom bir sistemdir. Bu sistem, bir hata veya alarm anında sorunun bağlamını otomatik olarak analiz eden bu hatayı teknik dokümanlardaki çözüm prosedürleriyle anlamsal olarak eşleştiren akıllı bir yöntem ile ilgilidir. Buluş, bu dinamik kök neden analizi sürecini, insan onayına sunulan otonom iyileştirme (veri tabanı güncelleme, konfigürasyon değişikliği veya yeniden işleme) aksiyonları ile birleştirir. Böylece hem hataların çözüm süresini (MTTR) minimize eden hem de operasyonel ekiplerin üzerindeki manuel analiz yükünü ortadan kaldıran, kendi kendini onaran (self-healing) bir veri işleme platformu sağlamaktadır. The invention is an AI-powered autonomous system that enables the management of complex data anomalies and processing errors (e.g., rejected CDR records) occurring in telecommunications mediation processes without the need for manual intervention. This system involves an intelligent method that automatically analyzes the context of an error or alarm and semantically matches the error with the solution procedures in the technical documentation. The invention combines this dynamic root cause analysis process with autonomous remediation actions (database updates, configuration changes, or reprocessing) that are subject to human approval. This provides a self-healing data processing platform that minimizes the time to resolve errors (MTTR) and eliminates the burden of manual analysis on operational teams.","assignee":"Vodafone Teknoloji̇ Hi̇zmetleri̇ Anoni̇m Şi̇rketi̇","inventors":["İlbey Özdemi̇r Zeren","Koçak İrem","Leloğlu Doğukan"],"publication_date":"2026-03-23","filing_date":"2025-12-24","priority_date":"2025-12-24","cpc_codes":["H","H04","H04L","H04L41/00","G","G06","G06F","G06F17/00","G","G06","G06N","G06N20/00"],"country":"TR","kind":"application","source_url":"https://patents.google.com/patent/TR2025021656A1/en"},{"publication_number":"KR20260039914A","title":"METHOD and APPARATUS for PROVIDING PERSONAL ARTIFICIAL INTELLIGENCE APPLICATION PACKS","abstract":"본 발명의 요약은 아래와 같다. ## 1. 전체 개요 본 발명은 **사용자 단말로부터 입력된 subject가 P·AI 시스템을 통해 처리되고, 외부 AI 서버 및 내부 학습·인지 구조를 거쳐 가공된 결과(return)가 다시 사용자에게 제공되는 전체 처리 흐름**을 나타낸다. 특히, 본 구조는 **다중 AI 연동, 인지축 기반 분석, 학습 DB 피드백**을 포함하는 순환형 지능처리 구조를 특징으로 한다. ## 2. 사용자 입력 및 인터페이스 단계 사용자는 **개인 컴퓨터 또는 모바일 단말**을 통해 subject를 입력한다. 해당 subject는 **P·AI 인터페이스[100]**로 전달되며, 인터페이스는 입력을 정규화하여 **P·AI서버[200]**로 전달한다. ## 3. 서버 분기 및 외부 AI 연동 P·AI 서버[200]는 수신된 subject를 내부 처리 흐름에 따라 분기한다. * 일부 subject는 **P·AI 전송부[300]**를 통해 **외부 AI 서버**로 전달되어 추가적인 분석 또는 응답 생성을 수행한다. * 외부 AI 서버는 처리 결과(response)를 다시 **P·AI 수신부[400]**로 반환한다. 이로써 P·AI 시스템은 **복수의 외부 AI로부터 응답을 수집할 수 있는 구조**를 가진다. ## 4. 인지 처리 및 인지축 분석 수신된 response는 **인지서버[500]**로 전달되어 인지 처리 단계에 진입한다. 인지서버는 응답을 **복수의 인지축(예: 人, 時, 物, 空, 法, 經, 技, 文)** 기준으로 분석하고, 각 인지축에 대응하는 **인지축수**를 산출한다. 이 인지축수는 단순 응답이 아닌, **의미·맥락·목적성을 반영한 정량적 인지 지표**로 작동한다. ## 5. 학습 DB 및 ML 구조 연동 산출된 인지축수는 **학습 DB**로 전달되어 누적되며, 해당 데이터는 **P·AI ML부**에서 학습 데이터로 활용된다. * 학습 DB는 인지축수에 기반한 **경험 축적 구조**를 형성한다. * ML부는 인지축 간 상관관계 및 패턴을 학습하여 이후 응답 처리에 반영한다. 이 과정은 **인지 → 학습 → 인지 고도화**로 이어지는 순환 구조를 형성한다. ## 6. 취득·가공 및 최종 결과 생성 학습 결과는 **취득서버[600]**로 전달되어 취득 값(return)을 산출한다. 이 취득 값은 다시 **가공서버[700]**에서 목적·subject·사용자 맥락에 맞게 가공된다. 가공된 결과(return)는 P·AI 서버를 거쳐 **P·AI 인터페이스[100]**로 전달되고, 최종적으로 **사용자 단말(개인 컴퓨터 또는 모바일)**에 출력된다. ## 7. 구조적 특징 요약 본 시스템은 다음과 같은 특징을 가진다. 1. **다중 AI 연동 구조** * 내부 AI + 외부 AI 서버의 병렬 활용 2. **인지축 기반 분석** * 단순 응답이 아닌 의미 중심 처리 3. **학습 DB 피드백 루프** * 응답 결과가 다시 학습에 반영되는 순환 구조 4. **개인·상황 적응형 return 생성** * subject 및 사용자 맥락을 반영한 가공 결과 제공 ## 8. 한 문장 요약 (초압축) 본 발명은 사용자의 subject가 P·AI 시스템에 입력되어 외부 AI 연동, 인지축 분석, 학습 DB 기반 처리 및 취득·가공 과정을 거쳐 다시 사용자에게 맞춤형 결과로 반환되는 지능형 처리 흐름을 나타낸다. 이를 시각적 다이아 그램으로 축약하면 아래 그림과 같다 The summary of the present invention is as follows. ## 1. Overall Overview The present invention represents an overall processing flow in which a subject input from a user terminal is processed through a P·AI system, and a processed result (return) is provided back to the user after passing through an external AI server and an internal learning and cognitive structure. In particular, this structure features a circular intelligent processing structure that includes **multiple AI integration, cognitive axis-based analysis, and learning DB feedback**. ## 2. User Input and Interface Phase The user enters the subject via a **personal computer or mobile device**. The subject is passed to the **P·AI interface[100]**, and the interface normalizes the input and passes it to the **P·AI server[200]**. ## 3. Server Branching and External AI Integration The P·AI server [200] branches the received subject according to the internal processing flow. Some subjects are transmitted to an external AI server via the P·AI transmission unit [300] to perform additional analysis or response generation. The external AI server returns the processing result (response) back to the **P·AI receiver[400]**. Thus, the P·AI system has a structure capable of collecting responses from multiple external AIs. ## 4. Cognitive Processing and Cognitive Axis Analysis The received response is forwarded to the **cognition server[500]** and enters the recognition processing phase. The cognitive server analyzes the response based on **multiple cognitive axes (e.g., person, time, thing, space, law, scripture, technique, text)** and calculates the **number of cognitive axes** corresponding to each cognitive axis. This cognitive axis functions not as a simple response, but as a **quantitative cognitive indicator reflecting meaning, context, and purpose**. ## 5. Integration of Training DB and ML Structure The calculated cognitive axes are transferred to the **training DB** and accumulated, This data is used as training data in the **P·AI ML Department**. The learning DB forms an **experience accumulation structure** based on cognitive axes. * The ML unit learns correlations and patterns between cognitive axes and reflects them in subsequent response processing. This process forms a cyclical structure leading to **cognition → learning → cognitive advancement**. ## 6. Acquisition, Processing, and Generation of Final Results The learning results are transmitted to the **acquisition server [600]** to calculate the return value. This return value is then processed at the **processing server [700]** to suit the purpose, subject, and user context. The processed result (return) is transmitted to the **P·AI interface[100]** via the P·AI server and finally output to the **user terminal (personal computer or mobile)**. ## 7. Summary of Structural Features This system has the following features. 1. **Multi-AI Integration Structure** * Parallel utilization of internal AI + external AI servers 2. **Cognitive Axis-Based Analysis** Semantic-centered processing rather than simple responses 3. **Learning DB Feedback Loop** A cyclic structure where response results are reflected back into learning 4. **Generating Individual/Context-Adaptive Returns** * Provides processed results reflecting the subject and user context ## 8. One-Sentence Summary (Ultra-Compressed) The present invention describes an intelligent processing flow in which a user's subject is input into a P·AI system, undergoes external AI integration, cognitive axis analysis, and processing based on a learning DB, and is returned to the user as a customized result. This is summarized in a visual diagram as shown in the figure below.","assignee":"최석진; 윤영배","inventors":["최석진","윤영배"],"publication_date":"2026-03-23","filing_date":"2025-12-22","priority_date":"2025-12-22","cpc_codes":["G","G06","G06F","G06F8/00","G06F8/30","G06F8/36","G","G06","G06F","G06F8/00","G06F8/30","G06F8/31","G06F8/315","G","G06","G06F","G06F8/00","G06F8/30","G06F8/38","G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/451","G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260039914A/en"},{"publication_number":"TR2025020149U5","title":"AN AI-POWERED DECISION ENGINE THAT GENERATES COSMETIC, SKIN CARE, AND DERMAL FORMULA PARAMETERS BASED ON USER SKIN DATA AND PREFERENCES.","abstract":"ÖZET Bu buluş; kullanıcıya ait cilt verileri ve kullanıcı tercihlerini çok değişkenli olarak değerlendirerek, kozmetik-cilt bakımı veya dermal ürünlerin üretiminde kullanılabilir üretim parametreleri oluşturan yapay zekâ destekli bir karar motoruna ilişkindir. Buluş kapsamında geliştirilen karar motoru; kullanıcıdan elde edilen yaş, cinsiyet, beslenme alışkanlıkları ve benzeri kişisel parametreler ile mobil cihaz kamerası aracılığıyla elde edilen cilt yüzeyine ait görsel verileri birlikte işleyen bir veri değerlendirme altyapısı içermektedir. Söz konusu veriler, sistem bünyesinde yer alan yapay zekâ ve makine öğrenmesi tabanlı karar mekanizması tarafından analiz edilerek kullanıcıya özgü üretim parametre setleri oluşturulmaktadır. Karar motoru, kullanıcı tarafından seçilen doğal, sentetik veya yarı sentetik içerik tercihleri ile cilt ihtiyaçlarını birlikte dikkate alacak şekilde yapılandırılmış olup, oluşturulan parametre setleri üretim sistemlerinde doğrudan kullanılabilecek nitelikte çıktılar oluşturmaktadır. Bu çıktılar; içerik sınıfları, oransal aralıklar ve üretim süreçlerine aktarılabilir veri yapıları şeklinde tanımlanmaktadır. Bu sayede, kişiye özel cilt bakım ürünlerinin formülasyon süreçleri otomatikleştirilmekte, manuel değerlendirme ihtiyacı azaltılmakta ve ölçeklenebilir, standartlaştırılmış bir üretim altyapısı oluşturulmaktadır. Buluş, cilt bakımı alanında kişiselleştirilmiş ve dinamik ürün üretimine yönelik yenilikçi bir teknik çözüm sunmaktadır. ABSTRACT This invention relates to an artificial intelligence-powered decision engine that generates production parameters usable in the production of cosmetic-skincare or dermal products by evaluating user skin data and user preferences in a multivariate manner. The decision engine developed within the scope of the invention includes a data evaluation infrastructure that processes personal parameters such as age, gender, and dietary habits obtained from the user together with visual data of the skin surface obtained via a mobile device camera. This data is analyzed by the artificial intelligence and machine learning-based decision mechanism within the system, and user-specific production parameter sets are created. The decision engine is structured to take into account the user's preferences for natural, synthetic, or semi-synthetic ingredients together with skin needs, and the generated parameter sets create outputs that can be directly used in production systems. These outputs are defined as ingredient classes, proportional ranges, and data structures that can be transferred to production processes. In this way, the formulation processes of personalized skincare products are automated, the need for manual evaluation is reduced, and a scalable, standardized production infrastructure is created. The invention offers an innovative technical solution for the production of personalized and dynamic products in the field of skincare.","assignee":"Bozan Murat","inventors":["Bozan Murat"],"publication_date":"2026-03-23","filing_date":"2025-12-13","priority_date":"2025-12-13","cpc_codes":["G","G06","G06N","G06N20/00"],"country":"TR","kind":"application","source_url":"https://patents.google.com/patent/TR2025020149U5/en"},{"publication_number":"CN121558358B","title":"A rolling bearing fault diagnosis method based on acoustic framing","abstract":"The invention discloses a rolling bearing fault diagnosis method based on acoustic framing. Firstly, hilbert transformation is carried out on acquired rolling bearing acoustic signals, and a double-channel analysis representation containing instantaneous amplitude and instantaneous phase information is constructed. And then, dividing the analytic signal into short-time frame fragments according to preset frame length and step length by an acoustic framing embedding module, and generating a frame-level embedding vector with unified dimension by linear mapping, so as to highlight local transient impact characteristics and effectively compress the sequence length. And (3) performing time sequence feature extraction by adopting an improved lightweight transducer encoder, wherein the encoder combines a low-dimensional multi-head self-attention mechanism with a feed-forward network based on SiLU activation and depth convolution so as to model a cross-frame global dependence and a short-range local structure at the same time, and finally, completing intelligent recognition of multiple types of faults of the rolling bearing through a global feature vector of a category token.","assignee":"Anhui University","inventors":["刘方","彭麟昊","王越悦","陆昂","程李","韦志清","陈硕","刘永斌"],"publication_date":"2026-03-20","filing_date":"2026-01-26","priority_date":"2026-01-26","cpc_codes":["G","G01","G01M","G01M13/00","G01M13/04","G01M13/045","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2131","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121558358B/en"},{"publication_number":"CN121565500B","title":"Clinical decision-making interaction methods and systems based on structured evidence reasoning","abstract":"The invention discloses a clinical decision interaction method and a system based on structured evidence reasoning, and relates to the technical field of computer technology and medical informatization. The method comprises the steps of obtaining user input query, carrying out mixed entity identification from the user input query based on an entity identification model and a rule matching algorithm to obtain a candidate entity set, carrying out conflict resolution and entity standardization processing on the candidate entity set to obtain a core entity, carrying out multi-granularity intention identification based on confidence calculation between the core entity and a classification system of medical exclusive intention to obtain target intention, carrying out screening query on a pre-constructed structured knowledge graph according to the core entity and the target intention to obtain a corresponding structured knowledge segment, generating a prompt word based on the structured knowledge segment, the user input query, instruction constraint and multiple verification, inputting the prompt word into a large language model, and outputting an answer. The invention can provide more accurate and reliable dialogue service for the medical field.","assignee":"Hainan University","inventors":["毋媛媛","黄无恙","黄梦醒","冯子凯"],"publication_date":"2026-03-20","filing_date":"2026-01-22","priority_date":"2026-01-22","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/70","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","Y","Y02","Y02A","Y02A90/00","Y02A90/10"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121565500B/en"},{"publication_number":"CN121544827B","title":"A method and system for intelligent identification of adverse geological conditions in tunnels based on ensemble learning","abstract":"The invention belongs to the field of geophysical exploration and tunnel engineering, and provides an intelligent tunnel bad geological identification method and system based on ensemble learning, which are used for constructing a three-dimensional geological model containing real geological information and randomly setting bad geological bodies of different categories; imaging the three-dimensional geologic model by different detection methods, extracting geologic feature parameters, storing the three-dimensional geologic model, imaging results and geologic feature parameters in a correlated way, extracting a multi-source feature map representing the position and the form of bad geologic features, performing deep fusion on the multi-source feature map and the geologic feature parameters to form a comprehensive feature vector reflecting the space form and the geologic attribute of the bad geologic features, training the integrated learning model by using the comprehensive feature vector, and processing target detection data by using the trained integrated learning model to obtain the identification result of the bad geologic features. The method can realize intelligent classification and prediction of the bad geological category and the risk level.","assignee":"Shandong University","inventors":["李术才","刘斌","袁伟","任玉晓","杨森林","李开元"],"publication_date":"2026-03-20","filing_date":"2026-01-19","priority_date":"2026-01-19","cpc_codes":["G","G06","G06T","G06T17/00","G06T17/05","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121544827B/en"},{"publication_number":"CN121524031B","title":"A Code Vulnerability Detection Method and System Based on Diffusion Suppression and Contrast Learning","abstract":"本发明属于代码检测领域，提供了一种基于扩散抑噪与对比学习的代码漏洞检测方法及系统，包括：基于源代码生成抽象语法树以及初始节点特征向量；将抽象语法树分解得到候选子树池，对每个候选子树中各初始节点特征向量进行局部编码得到候选子树向量；根据候选子树池构建子树图，对子树图中每个图节点的候选子树向量进行第一阶段扩散抑噪，得到去噪后的候选子树向量集合；逐一选取候选子树，得到因果子树集合；根据因果子树集合构建集合图，对因果子树集合进行第二阶段扩散抑噪，得到因果样本级集合向量；根据因果样本级集合向量进行漏洞检测，得到漏洞检测结果。本发明在保证源代码语义完整性的前提下，提升漏洞检测的准确性、鲁棒性及解释性。 This invention belongs to the field of code detection and provides a code vulnerability detection method and system based on diffusion denoising and contrastive learning. The method includes: generating an abstract syntax tree and initial node feature vectors based on the source code; decomposing the abstract syntax tree to obtain a candidate subtree pool, and locally encoding the feature vectors of each initial node in each candidate subtree to obtain candidate subtree vectors; constructing a subtree graph based on the candidate subtree pool, and performing a first-stage diffusion denoising on the candidate subtree vectors of each graph node in the subtree graph to obtain a denoised candidate subtree vector set; selecting candidate subtrees one by one to obtain a causal subtree set; constructing a set graph based on the causal subtree set, and performing a second-stage diffusion denoising on the causal subtree set to obtain a causal sample-level set vector; and performing vulnerability detection based on the causal sample-level set vector to obtain the vulnerability detection result. This invention improves the accuracy, robustness, and interpretability of vulnerability detection while ensuring the semantic integrity of the source code.","assignee":"Qilu University of Technology","inventors":["赵大伟","赵成晓","李鑫","徐丽娟","贾文琪","张则宇","宋上仁","仝丰华"],"publication_date":"2026-03-20","filing_date":"2026-01-16","priority_date":"2026-01-16","cpc_codes":["G","G06","G06F","G06F11/00","G06F11/36","G06F11/3604","G06F11/3608","G","G06","G06F","G06F11/00","G06F11/36","G06F11/3668","G","G06","G06N","G06N20/00","G","G06","G06N","G06N5/00","G06N5/01"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121524031B/en"},{"publication_number":"CN121505584B","title":"A method, device, equipment and medium for product identification in unmanned vending machines","abstract":"The invention relates to a commodity identification method, device, equipment and medium of an unmanned container, which belong to the technical field of deep learning, wherein the commodity identification method of the unmanned container acquires video data of a user for taking a commodity through a multi-view camera of the unmanned container, detects each frame of image data in the video data based on a commodity image detection network to obtain a commodity image, and identifies the commodity image based on a built MDFM network to obtain a commodity identification result, wherein the MDFM network comprises a multi-scale feature fusion module, a parallel transducer module, a Mamba module, a frequency domain processing module and a detection module, and the accuracy of commodity identification is improved.","assignee":"Wuhan Technical College of Communications","inventors":["肖强","姜凯","黄玮"],"publication_date":"2026-03-20","filing_date":"2026-01-14","priority_date":"2026-01-14","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/60","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121505584B/en"},{"publication_number":"CN121506167B","title":"An array speech enhancement method and system based on Hippo optimization and Mamba","abstract":"The invention provides an array voice enhancement method and system based on river horse optimization and Mamba, which belong to the technical field of audio signal processing and microphone arrays, and the invention utilizes a river horse optimization algorithm to perform robustness optimization on physical level on geometric configuration of a microphone array with the aim of maximizing white noise gain so as to obtain an optimal array layout; the optimal array is used for collecting multichannel noisy speech signals, an end-to-end speech enhancement model based on Mamba state space model is constructed, and the network directly maps the time-frequency characteristics of the multichannel noisy speech into the time-frequency mask of single-channel pure speech, so that efficient noise suppression and speech enhancement are realized in a dynamic environment with unknown sound source direction. The method solves the problems of poor robustness and performance degradation of the traditional method in a complex acoustic environment, remarkably improves the voice quality and the intelligibility, and is particularly suitable for scenes with low signal to noise ratio.","assignee":"Tianjin Polytechnic University","inventors":["刘意","闫铭熙","刘恒玮","孔德辉","王福宇","马浩然"],"publication_date":"2026-03-20","filing_date":"2026-01-13","priority_date":"2026-01-13","cpc_codes":["G","G10","G10L","G10L21/00","G10L21/02","G10L21/0208","G10L21/0216","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G10","G10L","G10L21/00","G10L21/02","G10L21/0316","G10L21/0324","G","G10","G10L","G10L21/00","G10L21/02","G10L21/0208","G10L21/0216","G10L2021/02161","G10L2021/02166"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121506167B/en"},{"publication_number":"CN121508596B","title":"A Sensing-Assisted Beam Prediction Method Based on Multimodal Spatiotemporal Fusion","abstract":"The invention discloses a perception auxiliary beam prediction method based on multi-mode space-time fusion, which uses RGB images, laser radar point clouds and millimeter wave radar signals as multi-mode perception inputs to carry out millimeter wave communication beam forming prediction, integrates complementary information of multi-mode perception in the beam forming prediction through cross-mode space fusion, extracts dynamic perception of historical time sequence information on a communication target through time sequence fusion coding, and can realize accurate beam forming prediction on fast moving targets such as vehicles. The cross-mode fusion and time sequence fusion part adopts a sequence coding technology based on a selection state space to replace a sequence coding technology based on an attention mechanism, and the calculation cost is smaller when the long sequence of the mode space fusion and the time sequence fusion is coded, so that the model training and reasoning speed is accelerated while the accuracy of beam forming prediction is ensured.","assignee":"Zhejiang University ZJU","inventors":["贺诗波","施振宇","顾超杰","钱滨"],"publication_date":"2026-03-20","filing_date":"2026-01-13","priority_date":"2026-01-13","cpc_codes":["H","H04","H04B","H04B7/00","H04B7/02","H04B7/04","H04B7/06","H04B7/0613","H04B7/0615","H04B7/0617","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/254","G06F18/256","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","H","H04","H04B","H04B7/00","H04B7/02","H04B7/04","H04B7/08","H04B7/0837","H04B7/0842","H04B7/086"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121508596B/en"},{"publication_number":"CN121479710B","title":"A model training method, system, device, and medium based on multimodal data","abstract":"The invention provides a model training method, a system, equipment and a medium based on multi-modal data, belonging to the technical field of artificial intelligence; the multi-modal data at least comprises original data of three modes of video, text and audio, the original feature vector is input into a multi-modal fusion model to be subjected to forward propagation processing to obtain a preliminary classification result, the association value of multi-modal fusion features of at least one level is calculated from the preliminary classification result in a reverse layer-by-layer mode, the multi-modal fusion features comprise initial fusion features, the multi-modal fusion features of the corresponding level are subjected to weighting processing based on the association value to obtain weighted fusion features, the weighted fusion features of each level are input into a corresponding subsequent level in the multi-modal fusion model again, and iterative training is performed to obtain a final multi-modal fusion model. The invention effectively improves the accuracy and generalization capability of the model.","assignee":"Wuhan College","inventors":["杜维","鲁圆圆","何璇","彭庆喜"],"publication_date":"2026-03-20","filing_date":"2026-01-09","priority_date":"2026-01-09","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121479710B/en"},{"publication_number":"CN121477213B","title":"A Multimodal Underwater Target Detection Method and System Based on Hierarchical Feature Alignment","abstract":"The invention relates to the technical field of underwater target detection, in particular to a multi-mode underwater target detection method and system based on layered feature alignment; the method comprises the steps of obtaining an underwater sonar image and an optical image, respectively extracting sonar and optical features through a double-flow backbone network, inhibiting background noise and enhancing key features through a local enhancement module, decomposing the features into low-frequency and high-frequency components through a layering alignment module, establishing a cross-modal feature corresponding relation through deformable convolution, generating multi-modal fusion features through a fusion module, and detecting targets through a detection head. The invention solves the problem of characteristic misalignment of the underwater sonar and the optical image caused by imaging principle difference, realizes self-adaptive cross-modal characteristic alignment and efficient fusion, and remarkably improves the accuracy and the robustness of underwater target detection.","assignee":"Guangdong Ocean University","inventors":["刘洺辛","谢俊杰","吴宇杰","郑志飞","杨万智","孔庆耀","盛昱秋","邹利兰"],"publication_date":"2026-03-20","filing_date":"2026-01-09","priority_date":"2026-01-09","cpc_codes":["G","G01","G01S","G01S15/00","G01S15/86","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06V","G06V10/00","G06V10/20","G06V10/25","G","G06","G06V","G06V10/00","G06V10/20","G06V10/30","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/40","G06V10/52","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/05","G","G06","G06V","G06V2201/00","G06V2201/07"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121477213B/en"},{"publication_number":"CN118709031B","title":"A method and system for diagnosing feature migration faults in a digital-analog linked rotor system","abstract":"The invention relates to a digital-analog linkage rotor system characteristic migration fault diagnosis method and system, which comprises the steps of introducing an unbalanced fault and an unbalanced fault into a constructed rotor system dynamics model to establish a rotor system fault dynamics differential equation and solve to obtain a fault displacement simulation signal, taking the fault displacement simulation signal as an input of a generator in an antagonism network through deep convolution of a constructed gradient punishment condition, inputting the obtained initial generation signal, a real signal and fault label information into a discriminator to obtain a generation signal integrating a mechanism characteristic and an actual mechanical characteristic, constructing a cross-working condition domain adaptation fault diagnosis model based on a characteristic migration learning theory, taking data of the generation signal integrating the mechanism characteristic and the actual mechanical characteristic as a source domain, taking other working condition data to be tested as a target domain, and training the cross-working condition domain adaptation fault diagnosis model to diagnose the rotor system cross-working condition fault.","assignee":"Beijing Information Science and Technology University","inventors":["王红军","马康"],"publication_date":"2026-03-20","filing_date":"2024-06-02","priority_date":"2024-06-02","cpc_codes":["G","G01","G01M","G01M13/00","G01M13/02","G","G01","G01M","G01M13/00","G","G01","G01M","G01M15/00","G","G01","G01M","G01M15/00","G01M15/02","G","G01","G01M","G01M15/00","G01M15/14","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2131","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN118709031B/en"},{"publication_number":"AU2026201596A1","title":"A method to mitigate allergen symptoms in a personalized and hyperlocal manner","abstract":"A system and method of determining an allergy impact profile of an individual are disclosed. The system and method may be employed to predict allergy impact environmental conditions may have on allergy symptoms of an individual and to recommend treatment of the individual in response to the predicted allergy impact.","assignee":"Kenvue Brands LLC","inventors":["Jennifer CALLAGHAN","Russell Gould","Grant HOU","Christina Lee","Jessica LIENERT","Matthew Machado","Thomas SHYR","Russel Walters"],"publication_date":"2026-03-19","filing_date":"2026-03-03","priority_date":"2019-09-24","cpc_codes":["G","G16","G16H","G16H20/00","G","G16","G16H","G16H20/00","G16H20/10","G","G01","G01D","G01D21/00","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0631","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/50","G","G16","G16H","G16H50/00","G16H50/70","A","A61","A61B","A61B5/00","A61B5/41","A61B5/411"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201596A1/en"},{"publication_number":"AU2026201529A1","title":"Estimating object properties using visual image data","abstract":"ESTIMATING OBJECT PROPERTIES USING VISUAL IMAGE DATA A system is comprised of one or more processors coupled to memory. The one or more processors are configured to receive image data based on an image captured using a camera of a vehicle and to utilize the image data as a basis of an input to a trained machine learning model to at least in part identify a distance of an object from the vehicle. The trained machine learning model has been trained using a training image and a correlated output of an emitting distance sensor. ESTIMATING OBJECT PROPERTIES USING VISUAL IMAGE DATA 20 26 20 15 29 27 F eb 2 02 6 2 0 2 6 2 0 1 5 2 9 2 7 F e b 2 0 2 6 s e n s o r .","assignee":"Tesla Inc","inventors":["Ashok Kumar ELLUSWAMY","James Anthony Musk","Swupnil Kumar Sahai"],"publication_date":"2026-03-19","filing_date":"2026-02-27","priority_date":"2019-02-19","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/251","G","G06","G06N","G06N20/00","G","G06","G06T","G06T7/00","G06T7/70","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/803","G","G06","G06V","G06V20/00","G06V20/50","G06V20/56","G06V20/58","G","G06","G06V","G06V20/00","G06V20/50","G06V20/56","G06V20/58","G06V20/584","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30248","G06T2207/30252","G06T2207/30261"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201529A1/en"},{"publication_number":"KR20260038878A","title":"A method for providing xr contents and xr device for providing xr contents","abstract":"본 발명은 적어도 두 개 이상의 디스플레이들을 포함하는 컨텐트 제공 디바이스에서 컨텐트를 제공하는 방법에 있어서, 제 1 컨텐트를 제 1 디스플레이 영역에 디스플레이하고, 제 2 컨텐트를 제 2 디스플레이 영역에 디스플레이하는 단계로서, 제 1 컨텐트는 네비게이션을 위한 XR 컨텐트를 포함하고, 제 2 컨텐트는 네비게이션과 관련된 정보를 포함하고, 사용자 입력 신호에 따라 상기 제 1 디스플레이 영역의 크기를 확장하고, 상기 제 2 디스플레이 영역의 크기를 축소하여 디스플레이하도록 제어할 수 있다. 또한, 컨텐트 제공 디바이스는 사용자 입력 신호에 따라 제 1 디스플레이 영역의 크기가 확장되면 제 2 디스플레이는 제 1 디스플레이 영역의 일부를 포함할 수 있다. 한편, 컨텐트 제공 디바이스는 제 1 디스플레이 영역 또는 제 2 디스플레이 영역 중 적어도 하나의 영역은 제 1 디스플레이 및 제 2 디스플레이에 포함되게 되고, 제 1 디스플레이 영역 및 제 2 디스플레이 영역은 서로 오버랩되지 않게 디스플레이 할 수 있다. The present invention relates to a method for providing content in a content providing device comprising at least two displays, wherein the method comprises the step of displaying a first content in a first display area and displaying a second content in a second display area, wherein the first content includes XR content for navigation and the second content includes information related to navigation, and the method may control the display by expanding the size of the first display area and reducing the size of the second display area according to a user input signal. Additionally, when the size of the first display area is expanded according to a user input signal, the second display may include a portion of the first display area. Meanwhile, the content providing device may display such that at least one of the first display area or the second display area is included in the first display and the second display, and the first display area and the second display area do not overlap each other.","assignee":"엘지전자 주식회사","inventors":["정두경"],"publication_date":"2026-03-19","filing_date":"2026-02-27","priority_date":"2018-12-06","cpc_codes":["G","G06","G06T","G06T11/00","G","G06","G06F","G06F3/00","G06F3/14","G06F3/1423","G06F3/1431","B","B60","B60K","B60K35/00","B60K35/10","B","B60","B60K","B60K35/00","B60K35/20","B60K35/28","B","B60","B60K","B60K35/00","B60K35/20","B60K35/29","G","G01","G01C","G01C21/00","G01C21/26","G01C21/34","G01C21/36","G01C21/3626","G01C21/3658","G","G01","G01C","G01C21/00","G01C21/26","G01C21/34","G01C21/36","G01C21/3664","G","G01","G01C","G01C21/00","G01C21/26","G01C21/34","G01C21/36","G01C21/3667","G01C21/367","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T11/00","G06T11/60","G06T11/65","G","G06","G06T","G06T15/00","G06T15/10","G06T15/20","G06T15/205","G","G06","G06T","G06T19/00","G06T19/006","G","G06","G06T","G06T19/00","G06T19/20","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G06","G06V","G06V20/00","G06V20/50","G06V20/56","G06V20/588","B","B60","B60K","B60K2360/00","B60K2360/16","B60K2360/164","B","B60","B60K","B60K2360/00","B60K2360/16","B60K2360/166","B","B60","B60K","B60K2360/00","B60K2360/16","B60K2360/177","B","B60","B60K","B60K2360/00","B60K2360/18","B60K2360/182","B","B60","B60K","B60K2360/00","B60K2360/20","B60K2360/31","B"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260038878A/en"},{"publication_number":"AU2026201476A1","title":"Digital medicine companion for treating and managing skin diseases","abstract":"DIGITAL MEDICINE COMPANION FOR TREATING AND MANAGING SKIN DISEASES A digital medicine companion for managing skin diseases (e.g., atopic dermatitis, psoriasis) may include patient wearable devices passively collecting patient data, patient user devices with healthcare applications for the patient to enter health related data, central analytics for flare 5 prediction and disease progress tracking, and a clinician dashboard. For flare prediction, a prediction tool may be trained using the ground truth of recorded flare occurrences for the trained model to predict whether observed scratch events may result in a flare. Upon predicting a likely flare, alert notifications may be generated, e.g., for a clinician and/or the patient. Furthermore, a baseline may be established based on the data passively gathered from the wearables and 10 actively gathered from the patient user devices. The continuously collected data may be compared against the established baseline. Upon detecting a significant deviation, alerts may be sent to the patient and/or the clinician. DIGITAL MEDICINE COMPANION FOR TREATING AND MANAGING SKIN DISEASES","assignee":"Pfizer Inc","inventors":["Yiorgos CHRISTAKIS","Robert Michael DAY","Junrui DI","Adam Fitzgerald","Dennis P. HANCOCK","Urs KERKMANN","Anthony LAMBROU","Fahimeh MAMASHLI","Timothy Mccarthy","Carrie Annalice NORTHCOTT","Joshua RAYSMAN","Felicia ZFIRA"],"publication_date":"2026-03-19","filing_date":"2026-02-26","priority_date":"2021-12-20","cpc_codes":["G","G06","G06N","G06N20/00","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H20/00","G","G16","G16H","G16H20/00","G16H20/10","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/50","G","G16","G16H","G16H50/00","G16H50/70","G","G16","G16H","G16H80/00"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201476A1/en"},{"publication_number":"AU2026201418A1","title":"Automated path-based recommendation for risk mitigation","abstract":"MARKED-UP COPY Systems and methods for automated path-based recommendation for risk mitigation are provided. An entity assessment server, responsive to a request for a recommendation for modifying a current risk assessment score of an entity to a target risk assessment score, accesses an input attribute vector for the entity and clusters of entities defined by historical attribute vectors. The entity assessment server assigns the input attribute vector to a particular cluster and determines a requirement on movement from a first point to a second point in a multi-dimensional space based on the statistics computed from the particular cluster. The first point corresponds to the current risk assessment score and the second point corresponds to the target risk assessment score. The entity assessment server computes an attribute-change vector so that a path defined by the attribute-change vector complies with the requirement and generates the recommendation from the attribute-change vector.","assignee":"Equifax Inc","inventors":["Mark Day","Lewis Jordan","Allan JOSHUA","Stephen Miller","Matthew Turner"],"publication_date":"2026-03-19","filing_date":"2026-02-25","priority_date":"2019-08-22","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q40/00","G06Q40/03"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201418A1/en"},{"publication_number":"KR20260038874A","title":"Time-Based Risk State Transition and Autonomous Shutdown System","abstract":"본 발명은 설비, 장비 또는 공간의 운영 상태를 관리하기 위한 것으로서, 분전반 또는 배전반의 전원선 등으로부터 비접촉 방식으로 수집되는 간접신호와, 진동, 음향 또는 생체 신호 등 서로 역할이 상이한 복수의 간접 계측 신호를 시간 축으로 분석하여 위험도를 산출하고, 상기 위험도의 시간적 지속성에 기초하여 상태전이 및 자율 정지를 수행하는 상태전이 기반 운영 시스템에 관한 것이다. 본 발명은 단일 신호의 순간적인 값 변화나 임계치 초과 여부에 의존하는 기존 방식과 달리, 위험 상태가 일정 시간 이상 지속되는 경우에만 상태전이를 수행하도록 구성된 시간 지속성 기반 판정 구조를 포함함으로써, 노이즈나 일시적 외란에 의한 오경보 및 불필요한 자동 정지를 구조적으로 억제하면서도, 실제 위험 상태의 초기 징후를 신뢰성 있게 포착할 수 있다. The present invention is for managing the operating status of facilities, equipment, or spaces, and, Indirect signals collected in a non-contact manner from power lines of distribution panels or switchboards, etc., and Calculate risk by analyzing multiple indirect measurement signals with different roles, such as vibration, acoustics, or biosignals, along the time axis, and This invention relates to a state transition-based operating system that performs state transitions and autonomous shutdowns based on the temporal persistence of the aforementioned risk level. Unlike existing methods that rely on instantaneous value changes of a single signal or whether a threshold is exceeded, the present invention, unlike, By including a time-persistence-based decision structure configured to perform a state transition only when a dangerous state persists for a certain period of time or longer, While structurally suppressing false alarms and unnecessary automatic shutdowns caused by noise or temporary disturbances, It can reliably capture early signs of actual danger.","assignee":"구교선; 구현우","inventors":["구교선","구현우"],"publication_date":"2026-03-19","filing_date":"2026-02-23","priority_date":"2026-02-23","cpc_codes":["G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0259","G05B23/0283","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0218","G05B23/0221","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0218","G05B23/0224","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0218","G05B23/0224","G05B23/0227","G05B23/0235","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260038874A/en"},{"publication_number":"KR20260038265A","title":"System and method for measuring state of color and thickness based on hyperspectral image analysis","abstract":"본 발명의 목적은 초분광 이미지의 분석 및 학습에 따라 피사체의 색상과, 이러한 색상의 변색에 따른 이상 상태 및 도료 두께에 따른 이상 상태를 실시간으로 분석하여 검출하는 초분광 이미지 분석에 기반한 컬러 및 두께 상태를 판정하는 시스템 및 방법을 제공하는 것이다. 상기 목적을 달성하기 위해, 본 발명에 따른 초분광 이미지 분석에 기반한 컬러 및 두께 상태를 판정하는 시스템은, 피사체의 초분광 이미지를 획득하는 초분광 카메라; 상기 초분광 이미지로부터 색상의 이상 상태를 판별하는 프로세서; 및 상기 프로세서에서 처리되는 이미지를 표시하는 표시부;를 포함하는 것을 특징으로 한다. The objective of the present invention is to provide a system and method for determining color and thickness conditions based on hyperspectral image analysis, which analyzes and detects the color of a subject, abnormal conditions resulting from discoloration of such color, and abnormal conditions resulting from paint thickness in real time according to the analysis and learning of hyperspectral images. To achieve the above objective, a system for determining color and thickness conditions based on hyperspectral image analysis according to the present invention is characterized by comprising: a hyperspectral camera for acquiring a hyperspectral image of a subject; a processor for determining an abnormal color condition from the hyperspectral image; and a display unit for displaying an image processed by the processor.","assignee":"(주)그린광학","inventors":["조현일","강태화","김태훈"],"publication_date":"2026-03-18","filing_date":"2026-03-11","priority_date":"2022-03-28","cpc_codes":["G","G01","G01J","G01J3/00","G01J3/28","G01J3/2823","G","G01","G01B","G01B11/00","G01B11/02","G01B11/06","G","G01","G01J","G01J3/00","G01J3/46","G","G01","G01N","G01N21/00","G01N21/17","G01N21/25","G01N21/27","G01N21/274","G","G01","G01N","G01N21/00","G01N21/17","G01N21/25","G01N21/31","G01N21/314","G01N21/3151","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G06T7/001","G","G06","G06T","G06T7/00","G06T7/50","G","G06","G06T","G06T7/00","G06T7/90","G","G01","G01N","G01N21/00","G01N21/17","G01N2021/1765","G01N2021/177","G01N2021/1776","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G01N2021/8854","G01N2021/888","G","G01","G01N","G01N2201/00","G01N2201/12","G01N2201/127"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260038265A/en"},{"publication_number":"KR20260036523A","title":"Method for analyzing state of electric power equipment","abstract":"본 발명의 전력설비 상태 판정 방법은 전력설비의 상태에 대한 진단데이터를 전처리하는 전처리부; 및 진단데이터를 다중 분석 알고리즘으로 분석하고 분석 결과의 정확도에 따라 진단데이터 각각의 가중치를 조정하여 통합하고, 진단데이터에서 이상을 탐지하고 오토인코더와 주축성분을 분석하여 전력설비의 상태를 도출하는 프로세서를 포함하는 것을 특징으로 한다. The power equipment condition determination method of the present invention is characterized by comprising: a preprocessing unit that preprocesses diagnostic data regarding the condition of power equipment; and a processor that analyzes the diagnostic data using a multi-analysis algorithm, adjusts and integrates the weights of each diagnostic data according to the accuracy of the analysis results, detects anomalies in the diagnostic data, and derives the condition of the power equipment by analyzing the autoencoder and the principal component.","assignee":"한국전력공사","inventors":["정진교","오중선","송호승"],"publication_date":"2026-03-17","filing_date":"2026-03-10","priority_date":"2022-08-02","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0218","G05B23/0221","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0259","G05B23/0275","G05B23/0281","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q50/00","G06Q50/06"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260036523A/en"},{"publication_number":"KR20260036520A","title":"Cooking device and operating method thereof","abstract":"본 개시의 실시 예에 따른 조리 기기는 조리실, 상기 조리실을 가열하는 가열부, 상기 조리실의 내부에 위치한 음식을 촬영하는 카메라, 사용자의 음성 명령을 수신하는 마이크로폰 및 상기 카메라를 통해 촬영된 음식 이미지로부터 상기 음식의 종류를 식별하고, 상기 음성 명령으로부터 상기 음식의 조리 상태를 나타내는 시각적 속성 정보를 획득하고, 상기 음식의 종류 및 상기 시각적 속성 정보에 매칭되는 조리 정보로 상기 음식을 조리하도록 상기 가열부를 제어하는 프로세서를 포함할 수 있다. A cooking device according to an embodiment of the present disclosure may include a cooking chamber, a heating unit for heating the cooking chamber, a camera for photographing food located inside the cooking chamber, a microphone for receiving a voice command from a user, and a processor for identifying the type of food from an image of food photographed through the camera, obtaining visual attribute information indicating the cooking state of the food from the voice command, and controlling the heating unit to cook the food using cooking information that matches the type of food and the visual attribute information.","assignee":"엘지전자 주식회사","inventors":["김재홍","정지욱","김효은","정연지","정형호","전혜정"],"publication_date":"2026-03-17","filing_date":"2026-03-10","priority_date":"2023-08-10","cpc_codes":["F","F24","F24C","F24C7/00","F24C7/08","F24C7/082","F24C7/085","A","A47","A47J","A47J36/00","A47J36/32","A47J36/321","A","A23","A23L","A23L5/00","A23L5/10","A23L5/15","A","A47","A47J","A47J27/00","A47J27/004","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06T","G06T7/00","G06T7/40","G","G06","G06T","G06T7/00","G06T7/60","G06T7/62","G","G06","G06T","G06T7/00","G06T7/90","G","G06","G06V","G06V10/00","G06V10/70","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/761","G","G06","G06V","G06V10/00","G06V10/70","G06V10/86","G","G06","G06V","G06V20/00","G06V20/50","G","G06","G06V","G06V20/00","G06V20/60","G06V20/68","H","H04","H04N","H04N23/00","H04N23/57","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30242"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260036520A/en"},{"publication_number":"KR20260037080A","title":"Method and system for language model learning","abstract":"본 발명은 언어 모델 학습 방법 및 시스템에 대한 것이다. 보다 구체적으로, 본 발명은 언어 모델의 언어 퇴화(Degeneration) 현상 없이 언어 모델을 학습할 수 있는 언어 모델 학습 방법 및 시스템에 대한 것이다. The present invention relates to a language model learning method and system. More specifically, the present invention relates to a language model learning method and system capable of learning a language model without the phenomenon of language degeneration.","assignee":"주식회사 Lg 경영개발원","inventors":["장영수","김건형","김병집","김유진","이홍락","이문태"],"publication_date":"2026-03-17","filing_date":"2026-02-27","priority_date":"2024-05-08","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260037080A/en"},{"publication_number":"KR20260036499A","title":"System and method for critical mineral trading based on Artificial Intelligence","abstract":"본 발명은 인공지능 기반 핵심광물 거래 시스템 및 그 방법을 제공하기 위한 것으로, 네트워크를 통해 상호 연결된 지능형 사용자 단말, 글로벌 광물 데이터베이스(DB) 서버, 광물생산 및 가공현장 서버, 및 이들을 통합 관리하는 AI 핵심광물 통합 관제 서버를 포함하되, 상기 AI 핵심광물 통합 관제 서버는 빅데이터 분석을 통해 수급 위기 지수를 산출하고, 사용자의 모의 투자 데이터를 기반으로 신뢰도를 검증하며, 실시간 재고량 데이터와 거래 정보를 연동하여 현장의 채굴 및 가공 설비에 대한 물리적 제어 신호를 송출하는 AI 중앙 제어부를 포함하여 구성함으로서, 글로벌 광물 데이터베이스와 연동하여 핵심광물의 수급 위기를 예측하고, 이를 지리정보시스템 기반의 리스크 지도로 시각화하며, 사용자 활동 데이터를 분석하여 동적으로 신뢰도를 검증하고, 생성형 AI를 활용해 상품 콘텐츠를 자동 저작하며, 거래 체결 데이터와 연동하여 광물 생산 및 가공 현장의 설비를 자율적으로 제어할 수 있다. The present invention provides an AI-based core mineral trading system and method, comprising an intelligent user terminal interconnected via a network, a global mineral database (DB) server, a mineral production and processing site server, and an AI core mineral integrated control server that manages them. The AI core mineral integrated control server is configured to include an AI central control unit that calculates a supply and demand crisis index through big data analysis, verifies reliability based on user simulated investment data, and transmits physical control signals to mining and processing facilities at the site by linking real-time inventory data and transaction information. By doing so, it is possible to predict supply and demand crises of core minerals by linking with the global mineral database, visualize them as a risk map based on a geographic information system, dynamically verify reliability by analyzing user activity data, automatically author product content using generative AI, and autonomously control facilities at mineral production and processing sites by linking with transaction execution data.","assignee":"정병렬","inventors":["정병렬"],"publication_date":"2026-03-17","filing_date":"2026-02-26","priority_date":"2026-02-26","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0613","G","G06","G06F","G06F16/00","G06F16/90","G06F16/907","G06F16/909","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/951","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087","G06Q10/0877","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0202","G","G06","G06Q","G06Q40/00","G06Q40/04","G06Q40/042","G06Q40/0421","G","G06","G06Q","G06Q50/00","G06Q50/02","G","G06","G06Q","G06Q50/00","G06Q50/04","G","G06","G06Q","G06Q50/00","G06Q50/10"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260036499A/en"},{"publication_number":"CN121504958B","title":"Medical image segmentation method based on multi-scale window alignment iterative fusion","abstract":"The application belongs to the technical field of medical image processing, and discloses a medical image segmentation method based on multi-scale window alignment iterative fusion, which comprises the steps of extracting a plurality of feature images with different scales from an input medical image by using an encoder; the method comprises the steps of performing iterative fusion on a plurality of feature graphs with different scales by utilizing a multi-scale window alignment fusion module to generate enhanced multi-scale features, decoding the enhanced multi-scale features by utilizing a receptive field attention decoder and combining jump connection features from an encoder to generate a plurality of stage segmentation feature graphs, fusing low-stage features of the encoder and high-stage features of the decoder by utilizing a spatial transposition fusion attention module to compensate information loss in a decoding process to generate a compensation feature graph, and aggregating the multi-stage segmentation feature graphs and the compensation feature graph to generate a final medical image segmentation result. The method effectively enhances the capturing, fusing and refining capability of the multi-scale features.","assignee":"Jiangxi Normal University","inventors":["罗勇","吴晗","王美云","徐磊","易玉根","唐权华"],"publication_date":"2026-03-17","filing_date":"2026-01-14","priority_date":"2026-01-14","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/40","G06V10/52","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121504958B/en"},{"publication_number":"CN121500044B","title":"Partial discharge identification method based on ultraviolet sensor array","abstract":"The invention discloses a partial discharge identification method based on an ultraviolet sensor array, and relates to the technical field of high-voltage electrical equipment monitoring. The method comprises the following steps of a, arranging 9 ultraviolet photodetectors arranged in a 3X3 matrix to form a sensor array, and independently collecting ultraviolet intensity signals generated by partial discharge by each sensing unit. According to the invention, the spatial distribution information of partial discharge is obtained by arranging the ultraviolet sensor array, the limitation of single-point detection is broken through, the anti-interference capability of the system in a complex electromagnetic environment is enhanced, the discriminant features are automatically extracted from the spatial intensity relation by utilizing the self-adaptive optimized feature transformation function, the defect that the traditional method depends on manual experience is overcome, the validity of the features and the generalized recognition capability of modes are improved, and the adaptability and the data stability of the system to random noise in actual working conditions are further improved by combining preprocessing and data enhancement strategies.","assignee":"Hefei Meigallium Sensing Technology Co ltd","inventors":["王子韩","刘云鹏","孙剑文","刘煦"],"publication_date":"2026-03-17","filing_date":"2026-01-12","priority_date":"2026-01-12","cpc_codes":["G","G01","G01R","G01R31/00","G01R31/12","G01R31/1218","G","G01","G01R","G01R31/00","G01R31/12","G01R31/1227","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/217","G06F18/2178","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121500044B/en"},{"publication_number":"CN121482404B","title":"Multi-expert collaborative medical image segmentation method based on multi-scale information fusion","abstract":"A multi-expert collaborative medical image segmentation method based on multi-scale information fusion belongs to the field of computer vision. The method comprises the steps of respectively capturing global semantic information and local detail information of a medical image by constructing a dual-path encoder, designing cross-scale interaction fusion to realize efficient information exchange among different scale features, constructing a single-path decoder module, and gradually recovering the space detail information to generate a segmentation mask. In addition, the comprehensive supervision method training of the deep hierarchical supervision loss function is adopted. Experimental results show that the method is superior to the current mainstream method on a plurality of medical image data sets, the segmentation precision and the robustness are remarkably improved, and a new technical scheme is provided for the field of medical image segmentation.","assignee":"Dalian University of Technology","inventors":["郑国良","李朋"],"publication_date":"2026-03-17","filing_date":"2026-01-09","priority_date":"2026-01-09","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/20","G06V10/26","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121482404B/en"},{"publication_number":"CN121457543B","title":"Large language model back-end implementation system based on reasoning service","abstract":"本发明提供一种基于推理服务的大语言模型后端实现系统，属于计算机技术领域，所述系统包括硬件模块、推理框架模块、大语言模型后端模块和推理服务模块，其中，所述硬件模块包括CPU、GPU、NPU计算资源；所述推理框架模块包括开源大语言模型框架和通用第三方自研推理框架；所述大语言模型后端模块包括通用模块和大语言模型实例模块；所述推理服务模块包括vllm推理服务、llama.cpp服务或者自研推理服务。本发明多种推理服务通过大语言模型后端模块实现了不同推理框架推理，提高了开发者使用效率。 This invention provides a large language model backend implementation system based on inference services, belonging to the field of computer technology. The system includes a hardware module, an inference framework module, a large language model backend module, and an inference service module. The hardware module includes CPU, GPU, and NPU computing resources. The inference framework module includes open-source large language model frameworks and general-purpose third-party self-developed inference frameworks. The large language model backend module includes general modules and large language model instance modules. The inference service module includes VLLM inference services, llama.cpp services, or self-developed inference services. This invention implements various inference services through the large language model backend module, achieving inference with different inference frameworks and improving developer efficiency.","assignee":"Kirin Software Co Ltd","inventors":["宾泽民","刘意虎","张铎","张超","李真能","吴江","朱晨"],"publication_date":"2026-03-17","filing_date":"2026-01-07","priority_date":"2026-01-07","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/445","G06F9/44505","G06F9/4451","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5027","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121457543B/en"},{"publication_number":"CN121454571B","title":"Method and system for forecasting ionospheric delay by combining ConvLSTM and ViT fusion model","abstract":"The invention discloses a method for forecasting ionosphere delay by combining a ConvLSTM and ViT fusion model, which comprises the steps of obtaining single-moment TEC data provided by an ionosphere experience model in a target area, VTEC data and a mask layer obtained by a synchronous ground GNSS site, inputting the obtained data into the trained fusion model, and outputting a TEC forecasting result of a future preset period, wherein the fusion model combines a ConvLSTM model and a ViT model. According to the invention, local space-time details are captured through ConvLSTM, and the structure of capturing global dependence and multi-input fusion auxiliary information is combined with ViT, so that ionospheric delay information can be well predicted under the condition of active ionosphere.","assignee":"Wuhan University WHU","inventors":["胡茂华","朱锋"],"publication_date":"2026-03-17","filing_date":"2026-01-07","priority_date":"2026-01-07","cpc_codes":["G","G01","G01S","G01S19/00","G01S19/01","G01S19/13","G01S19/35","G01S19/37","G","G01","G01S","G01S19/00","G01S19/01","G01S19/03","G01S19/07","G01S19/072","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","Y","Y02","Y02A","Y02A90/00","Y02A90/10"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121454571B/en"},{"publication_number":"CN121479468B","title":"Method, system, device and storage medium for detecting faults of grinding machine","abstract":"The application discloses a method, a system, a device and a storage medium for detecting faults of a grinding machine, which are used for intelligently detecting the faults of the grinding machine under the condition of extreme data. The method for detecting the faults of the grinding machine comprises the steps of obtaining multi-mode data of a target grinding machine, preprocessing the multi-mode data by using a preset multi-mode fusion diagnosis model to obtain preprocessed multi-mode features, calculating confidence coefficient of the multi-mode features by using the multi-mode fusion diagnosis model, distributing weights to the features of each mode in the multi-mode features by using the multi-mode fusion diagnosis model according to a preset distribution mechanism and the confidence coefficient, carrying out feature fusion processing on the multi-mode features by using the multi-mode fusion diagnosis model on the basis of the distributed weights to obtain fusion features, inputting the fusion features into a classifier of the multi-mode fusion diagnosis model, and outputting fault classification diagnosis results of the target grinding machine.","assignee":"Guiyang Xianfeng Machine Tool Co ltd; Guizhou University","inventors":["魏建安","原一航","吴长城","杨佐德","张芝然"],"publication_date":"2026-03-17","filing_date":"2026-01-07","priority_date":"2026-01-07","cpc_codes":["G","G01","G01M","G01M99/00","G01M99/005","G","G01","G01D","G01D21/00","G01D21/02","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06F","G06F2123/00","G06F2123/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121479468B/en"},{"publication_number":"CO2026002985A2","title":"Technologies to leverage enhanced conversational bots in a contact center system","abstract":"RESUMEN Un método para aprovechar bots conversacionales mejorados en un sistema de centro de contacto de acuerdo con una modalidad incluye realizar destilación de conocimiento a través del aprendizaje automático para enseñar un modelo de inteligencia artificial del estudiante basado en un modelo de inteligencia artificial del docente y reducir un tamaño de un vocabulario multilingüe inicial, en donde el modelo de inteligencia artificial del estudiante incluye menos capas de incrustación de aprendizaje automático que el modelo de inteligencia artificial del docente, eliminando tókenes del vocabulario multilingüe inicial basado en una agrupación de lenguajes con similitudes lingüísticas para reducir el tamaño del vocabulario multilingüe inicial, analizar el texto de usuario de un usuario humano en uno o más tókenes, identificar índices de token asociados con el respectivo uno o más tókenes en un vocabulario multilingüe reducido, determinar valores de incrustación asociados con los índices de token identificados y generar una salida de incrustación multilingüe para el texto de usuario indicativa de una intención del usuario basada en los valores de incrustación usando aprendizaje automático. ABSTRACT A method for leveraging enhanced conversational bots in a contact center system according to a modality includes performing knowledge distillation through machine learning to teach a student AI model based on a teacher AI model and reduce the size of an initial multilingual vocabulary, wherein the student AI model includes fewer machine learning embedding layers than the teacher AI model, removing tokens from the initial multilingual vocabulary based on a grouping of languages with linguistic similarities to reduce the size of the initial multilingual vocabulary, analyzing the user text of a human user into one or more tokens, identifying token indices associated with the respective one or more tokens in a reduced multilingual vocabulary, determining embedding values associated with the identified token indices, and generating a multilingual embedding output for the user text indicative of a user intent based on the embedding values using machine learning.","assignee":"Genesys Cloud Services Inc","inventors":["Ramasubramanian Sundaram","Naresh Kumar Elluru","Pavan Kumar Buduguppa"],"publication_date":"2026-03-16","filing_date":"2026-03-12","priority_date":"2023-08-31","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G06F40/216","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06F","G06F40/00","G06F40/30","G06F40/35","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CO","kind":"application","source_url":"https://patents.google.com/patent/CO2026002985A2/en"},{"publication_number":"KR20260036264A","title":"Apparatus for updating an ocean and climate environment prediction model through feedback of real-time observation data","abstract":"전자 장치의 동작 방법이 개시된다. 본 개시에 따른 전자 장치의 동작 방법은, 일 해양 구역에 대한 복수의 환경 변수의 값을 실시간으로 획득하는 단계 및 다중 LSTM(Long-Short Term Memory) 구조 상에 실시간으로 획득된 복수의 환경 변수의 값을 입력하여, 시계열에 따른 복수의 환경 변수 각각의 값을 예측하는 단계를 포함한다. A method of operating an electronic device is disclosed. The method of operating an electronic device according to the present disclosure includes the step of acquiring values of a plurality of environmental variables for a marine area in real time, and the step of inputting the values of the plurality of environmental variables acquired in real time onto a multiple LSTM (Long-Short Term Memory) structure to predict the value of each of the plurality of environmental variables according to a time series.","assignee":"주식회사 시즈","inventors":["이미애","이호근","함동빈"],"publication_date":"2026-03-16","filing_date":"2026-03-06","priority_date":"2024-09-06","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G01","G01W","G01W1/00","G01W1/02","G","G01","G01W","G01W1/00","G01W1/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G01","G01W","G01W2201/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260036264A/en"},{"publication_number":"KR20260036235A","title":"Electronic apparatus for providing pay later service and method thereof","abstract":"본 개시는 아이템 판매 플랫폼과 관련된 제1 전자 장치에 의한 추후 결제 서비스 제공 방법에 관한 것으로, 상기 추후 결제 서비스 제공 방법은 사용자의 단말으로부터 상기 아이템에 대한 결제 요청을 획득하는 단계, 상기 플랫폼에서의 상기 사용자의 활동 정보를 기반으로 상기 사용자의 추후 결제 서비스 이용 가능 여부를 확인하는 단계 및 상기 사용자의 추후 결제 서비스 이용이 가능한 경우, 상기 결제 요청에 대응하여 상기 추후 결제 서비스에 대한 제1 정보를 제공하는 단계를 포함한다. The present disclosure relates to a method for providing a subsequent payment service by a first electronic device associated with an item sales platform, wherein the method for providing a subsequent payment service comprises the steps of: obtaining a payment request for the item from a user's terminal; checking whether the user is eligible to use the subsequent payment service based on the user's activity information on the platform; and, if the user is eligible to use the subsequent payment service, providing first information regarding the subsequent payment service in response to the payment request.","assignee":"쿠팡 주식회사","inventors":["임호현","궈샹 딩","임이랑","배미성","김진혁","예병욱","육민용","이주연","조선미"],"publication_date":"2026-03-16","filing_date":"2026-02-23","priority_date":"2021-12-06","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q20/00","G06Q20/02","G","G06","G06Q","G06Q20/00","G06Q20/08","G06Q20/12","G","G06","G06Q","G06Q20/00","G06Q20/22","G06Q20/24","G","G06","G06Q","G06Q20/00","G06Q20/38","G","G06","G06Q","G06Q20/00","G06Q20/38","G06Q20/389","G","G06","G06Q","G06Q20/00","G06Q20/38","G06Q20/40","G","G06","G06Q","G06Q20/00","G06Q20/38","G06Q20/40","G06Q20/401","G06Q20/4014","G","G06","G06Q","G06Q20/00","G06Q20/38","G06Q20/40","G06Q20/403","G","G06","G06Q","G06Q20/00","G06Q20/38","G06Q20/42","G","G06","G06Q","G06Q30/00","G06Q30/06","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260036235A/en"},{"publication_number":"BG5250U1","title":"Integrated medical platform with domain-restricted artificial intelligence","abstract":"The present utility model relates to an integrated medical platform with domain-restricted artificial intelligence, intended for application in the fields of medical information systems, digital healthcare, telemedicine, and first aid and medical support systems.The developed integrated digital platform comprises a module for input of medical informational content (1), connected to a module for processing and semantic representation of the information (2), which is connected to a domain-structured vector database (3).The domain-structured vector database (3) is functionally connected to a rights management module (4), which is connected to a module for retrieval and response generation using artificial intelligence (5).The module for retrieval and response generation using artificial intelligence (5) is connected to a device with an online client interface (6) and to a local offline client device (7). Both the online client interface device (6) and the local offline client device (7) are bidirectionally connected to a module containing reference medical values and evaluation algorithms (8).","assignee":"\"Прокон-М\" Еоод; Радков Кесов Георги","inventors":["Георги Кесов","Радков Кесов Георги"],"publication_date":"2026-03-16","filing_date":"2026-02-12","priority_date":"2026-02-12","cpc_codes":["G","G06","G06F","G06F12/00","G06F12/02","G","G06","G06N","G06N20/00","G","G16","G16H","G16H10/00","G","G16","G16H","G16H40/00","G","G16","G16H","G16H50/00"],"country":"BG","kind":"application","source_url":"https://patents.google.com/patent/BG5250U1/en"},{"publication_number":"KR20260035880A","title":"Electronic apparatus and method for predicting ocean and meteorological observation data by sequentially connecting a plurality of lstm models","abstract":"전자 장치의 동작 방법이 개시된다. 본 개시에 따른 전자 장치의 동작 방법은, 일 해양 구역에 대한 복수의 환경 변수의 값을 실시간으로 획득하는 단계 및 다중 LSTM(Long-Short Term Memory) 구조 상에 실시간으로 획득된 복수의 환경 변수의 값을 입력하여, 시계열에 따른 복수의 환경 변수 각각의 값을 예측하는 단계를 포함한다. A method of operating an electronic device is disclosed. The method of operating an electronic device according to the present disclosure includes the step of acquiring values of a plurality of environmental variables for a marine area in real time, and the step of inputting the values of the plurality of environmental variables acquired in real time onto a multiple LSTM (Long-Short Term Memory) structure to predict the value of each of the plurality of environmental variables according to a time series.","assignee":"주식회사 시즈","inventors":["이미애","이호근","함동빈"],"publication_date":"2026-03-13","filing_date":"2026-03-06","priority_date":"2024-09-06","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G01","G01W","G01W1/00","G01W1/02","G","G01","G01W","G01W1/00","G01W1/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G01","G01W","G01W2201/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260035880A/en"},{"publication_number":"KR20260035869A","title":"Device for Automated Curation and Dual-Path Injection of High-Quality AI Training Data Using Multi-Dimensional Interaction Weighting","abstract":"본 발명은 인공지능 학습용 고품질 데이터의 자동 큐레이션 및 이원화 주입 장치에 관한 것이다. 본 발명은 온라인 커뮤니티 내 사용자의 상호작용 데이터를 비용성 반응과 비비용성 반응으로 구분하여 다중 모달 벡터화하고, 평가자의 신뢰도 지수를 반영한 동적 가중치 연산을 통해 콘텐츠의 최종 품질 점수를 산출한다. 또한, 전체 데이터의 점수 분포를 실시간으로 분석하여 상위 기 설정된 퍼센타일(예: 5%)의 동적 임계값을 초과하는 최상위 골든 데이터만을 추출하는 적응형 임계치 설정 및 게이팅을 수행한다. 추출된 데이터는 이원화 주입 제어기를 통해 지식 강화를 위한 고차원 벡터 저장 장치(실시간 증분 인덱싱)와, 지능 강화를 위한 자체 도메인 특화 모델(배치 학습 및 샌드박스 무결성 검증 후 배포)로 각각 자동 분기되어 주입된다. 이를 통해 어뷰징 트래픽을 원천 차단한 고순도 학습 데이터를 확보하고, 실시간 검색 품질 향상과 자체 인공지능 모델의 무중단 진화를 동시에 달성하는 자생적 데이터 선순환 파이프라인을 제공한다. The present invention relates to an automatic curation and dual injection device for high-quality data for artificial intelligence training. The invention classifies user interaction data within an online community into cost-effective and non-cost-effective responses, performs multimodal vectorization, and calculates a final quality score of the content through dynamic weighting operations reflecting the evaluator's reliability index. Additionally, it performs adaptive threshold setting and gating by analyzing the score distribution of the entire data in real time to extract only the top \"golden data\" that exceeds a dynamic threshold of a pre-set upper percentile (e.g., 5%). The extracted data is automatically branched and injected through a dual injection controller into a high-dimensional vector storage device for knowledge enhancement (real-time incremental indexing) and a proprietary domain-specific model for intelligence enhancement (distribution after batch training and sandbox integrity verification). Through this, high-purity training data is secured by blocking abusive traffic at the source, and a self-sustaining data virtuous cycle pipeline is provided that simultaneously achieves real-time search quality improvement and uninterrupted evolution of the proprietary artificial intelligence model.","assignee":"김희봉","inventors":["김희봉"],"publication_date":"2026-03-13","filing_date":"2026-02-24","priority_date":"2026-02-24","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F21/00","G06F21/50","G06F21/52","G06F21/53","G","G06","G06F","G06F8/00","G06F8/60","G06F8/65","G","G06","G06F","G06F8/00","G06F8/70","G06F8/71"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260035869A/en"},{"publication_number":"KR20260035865A","title":"Apparatus and method for machine learning-based domain derivation","abstract":"본 개시는 IP(Internet Protocol) 주소와 연결된 도메인을 도출하는 전자 장치 및 그 방법에 관한 것으로, 상기 전자 장치는 외부와 통신을 수행하기 위한 통신부, 메모리 및 하나 이상의 코어를 포함하는 프로세서를 포함하고, 상기 프로세서는 HTML(Hypertext Markup Language) 소스를 기반으로 대상 IP 주소에 대응하는 복수의 도메인 정보들을 탐지하고, 상기 복수의 도메인 정보들과 관련된 입력 데이터 세트를 기계 학습이 수행된 복수의 모델들 각각에 대하여 입력하고, 상기 입력 데이터 세트를 기반으로 상기 복수의 모델들 각각으로부터 출력 값을 도출하고, 상기 복수의 모델들 각각에 대응하는 가중치를 산출하고, 상기 복수의 모델들 각각으로부터 도출된 출력 값 및 상기 복수의 모델들 각각에 대응하는 가중치를 기반으로 복수의 도메인 정보들 각각에 대응하는 최종 출력 값을 도출하고, 상기 복수의 도메인 정보들 각각에 대응하는 최종 출력 값을 비교하여 대표 도메인 정보를 선택하는 것을 특징으로 한다. The present disclosure relates to an electronic device and a method for deriving a domain associated with an IP (Internet Protocol) address, wherein the electronic device comprises a communication unit for performing communication with the outside, a memory, and a processor including one or more cores, wherein the processor detects a plurality of domain information corresponding to a target IP address based on an HTML (Hypertext Markup Language) source, inputs an input data set related to the plurality of domain information to each of a plurality of models on which machine learning has been performed, derives an output value from each of the plurality of models based on the input data set, calculates a weight corresponding to each of the plurality of models, derives a final output value corresponding to each of the plurality of domain information based on the output value derived from each of the plurality of models and the weight corresponding to each of the plurality of models, and selects a representative domain information by comparing the final output value corresponding to each of the plurality of domain information.","assignee":"주식회사 에이아이스페라","inventors":["강병탁","최동식"],"publication_date":"2026-03-13","filing_date":"2026-02-24","priority_date":"2023-11-15","cpc_codes":["G","G06","G06N","G06N20/00","G06N20/10","H","H04","H04L","H04L61/00","H04L61/45","G","G06","G06N","G06N20/00","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N5/00","G06N5/01","H","H04","H04L","H04L63/00","H04L63/16","H04L63/166","H","H04","H04L","H04L67/00","H04L67/01","H04L67/02","H","H04","H04L","H04L43/00","H04L43/02","H04L43/026","H","H04","H04L","H04L61/00","H04L61/45","H04L61/4505","H04L61/4511"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260035865A/en"},{"publication_number":"KR20260035857A","title":"Anomaly detection method and appratus using neural network","abstract":"뉴럴 네트워크를 이용하여 이상치를 탐지하는 전자 장치가 개시된다. 일 실시예에 따른 전자 장치는 메모리, 송수신기, 및 프로세서를 포함하고, 상기 프로세서는, 입력 이미지에 기초하여 산출된 심층 특징(feature) 및 상기 심층 특징의 평균에 기초하여 상기 입력 이미지에 대응하는 이상치를 탐지하도록 미리 학습된 뉴럴 네트워크 모델에 기초하여, 상기 입력 이미지의 이상치를 탐지할 수 있다. An electronic device for detecting outliers using a neural network is disclosed. An electronic device according to one embodiment includes a memory, a transceiver, and a processor, wherein the processor can detect outliers in an input image based on a neural network model that is pre-trained to detect outliers corresponding to the input image based on deep features calculated based on the input image and the average of said deep features.","assignee":"주식회사 엘로이랩","inventors":["유광선","윤혁"],"publication_date":"2026-03-13","filing_date":"2026-02-23","priority_date":"2022-05-23","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T7/00","G06T7/0002","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/7715","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260035857A/en"},{"publication_number":"KR20260035863A","title":"Method and system for managing produrct information","abstract":"본 개시는 상품 정보 관리 방법 및 그 시스템에 관한 것이다. 본 개시의 일 실시예들에 따른 상품 정보 관리 방법은, 제1 SKU에 대한 디스크립션 컨텐츠에 대한 시맨틱 분석(semantic analysis)을 수행하는 단계, 상기 시맨틱 분석의 결과를 이용하여, 상기 제1 SKU의 시맨틱 피처 셋(semantic feature set) -상기 시맨틱 피처 셋은 브랜드를 표현하는 제1 피처 및 상품을 표현하는 제2 피처를 포함하는 복수의 피처로 구성된- 을 추출하는 단계, 상기 제1 SKU의 추출된 상기 시맨틱 피처 셋을 이용하여, 상기 제1 SKU와 각각의 비교 대상 SKU 사이의 유사도를 산출하는 단계 및 상기 유사도에 기반하여 상기 비교 대상 SKU 중 적어도 하나의 SKU를 상기 제1 SKU에 대응하는 유사 SKU로 결정하는 단계를 포함할 수 있다. The present disclosure relates to a method for managing product information and a system thereof. A method for managing product information according to one embodiment of the present disclosure may include: performing a semantic analysis on description content for a first SKU; using the result of the semantic analysis, extracting a semantic feature set of the first SKU—the semantic feature set being composed of a plurality of features including a first feature representing a brand and a second feature representing a product—; using the extracted semantic feature set of the first SKU, calculating a similarity between the first SKU and each comparison target SKU; and determining at least one SKU among the comparison target SKUs as a similar SKU corresponding to the first SKU based on the similarity.","assignee":"쿠팡 주식회사","inventors":["리우위슈","푸구이얀","친밍","추이유","장광야오","웨이웨이","린젠워이","징위허","앙시후","최수경"],"publication_date":"2026-03-13","filing_date":"2026-02-23","priority_date":"2024-07-03","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F40/00","G06F40/20","G06F40/268","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260035863A/en"},{"publication_number":"KR20260035864A","title":"Device and Method for Providing Answer Generation and Content Curation Using Intelligent Parallel Processing Scheduler and High-Dimensional Latent Space Control","abstract":"본 발명은 인공지능 기반의 답변 생성 및 연관 콘텐츠 큐레이션 제공 장치에 관한 것이다. 본 장치는 사용자 질의를 수신하는 즉시 외부 AI 추론과 내부 벡터 검색으로 프로세스를 이원화하는 지능형 병렬 프로세싱 스케줄러 를 포함하며, 이를 통해 응답 지연 시간을 단축하고 고차원 잠재 공간 임베딩 엔진을 통해 검색 정밀도를 향상시킨다. The present invention relates to an AI-based answer generation and related content curation provision device. The device includes an intelligent parallel processing scheduler that bifurcates the process into external AI inference and internal vector search immediately upon receiving a user query, thereby reducing response latency and improving search precision through a high-dimensional latent space embedding engine.","assignee":"김희봉","inventors":["김희봉"],"publication_date":"2026-03-13","filing_date":"2026-02-23","priority_date":"2026-02-23","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3347","G","G06","G06F","G06F16/00","G06F16/20","G06F16/22","G06F16/2228","G06F16/2237","G","G06","G06F","G06F16/00","G06F16/30","G06F16/31","G06F16/316","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260035864A/en"},{"publication_number":"AU2026201380A1","title":"Generating videos using sequences of generative neural networks","abstract":"53 Methods, systems, and apparatus, including computer programs encoded on a computer storage medium. In one aspect, a method includes receiving a text prompt describing a scene; processing the text prompt using a text encoder neural network to generate a contextual embedding of the text prompt; and processing the contextual embedding using a sequence of generative neural networks to generate a final video depicting the scene. 53 20 26 20 13 80 24 F eb 2 02 6 2 0 2 6 2 0 1 3 8 0 2 4 F e b 2 0 2 6 5 3","assignee":"Google LLC","inventors":["William Chan","Jonathan HO","Chitwan SAHARIA","Tim SALIMANS","Jay Ha WHANG"],"publication_date":"2026-03-12","filing_date":"2026-02-24","priority_date":"2023-03-24","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06T","G06T3/00","G06T3/40","G06T3/4053"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201380A1/en"},{"publication_number":"AU2026201300A1","title":"Data processing system for generating predictions of cognitive outcome in patients","abstract":"A system for outputting a visual representation of a brain of a patient is configured to receive sensor data representing a behavior of a region of the brain of the patient. The system retrieves mapping data that maps a prediction value to the region. The prediction value is indicative of an effect on a behavior of the patient responsive to a treatment of the region, the mapping data being indexed to a patient identifier. The system receives, responsive to an application of a stimulation to the region, sensor data representing behavior of the region. The system executes a model that updates, based on the sensor data, the prediction value for the region. The system updates, responsive to executing the model, the mapping data by including the updated prediction value in the mapping data. The system outputs a visual representation of the updated mapping data comprising the updated prediction value.","assignee":"Carnegie Mellon University","inventors":["Hugo ANGULO-ORQUERA","Benjamin CHERNOFF","Bradford MAHON","Keith Parkins","Max SIMS"],"publication_date":"2026-03-12","filing_date":"2026-02-20","priority_date":"2018-11-30","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/70","A","A61","A61N","A61N1/00","A61N1/02","A61N1/04","A61N1/05","A61N1/0526","A61N1/0529","A","A61","A61B","A61B34/00","A61B34/10","A","A61","A61B","A61B5/00","A61B5/0033","A61B5/004","A61B5/0042","A","A61","A61B","A61B5/00","A61B5/0059","A61B5/0077","A","A61","A61B","A61B5/00","A61B5/05","A61B5/055","A","A61","A61B","A61B5/00","A61B5/24","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/369","A61B5/372","A","A61","A61B","A61B5/00","A61B5/40","A61B5/4058","A61B5/4064","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4848","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","A","A61","A61B","A61B5/00","A61B5/74","A61B5/742","A61B5/7425","A","A61","A61N","A61N1/00","A61N1/18","A61N1/32","A61N1/36","A61N1/36014","A","A61","A61N","A61N1/00","A61N1/18","A61N1/32","A61N1/36","A61N1/362","A61N1/37","G","G01","G01R","G01R33/00","G01R33/20","G01R33/44","G01R33/48","G01R33/4806","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H20/00","G16H20/30","G"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201300A1/en"},{"publication_number":"AU2026201319A1","title":"Allocating computing resources between model size and training data during training of a machine learning model","abstract":"Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a machine learning model to perform a machine learning task. In one aspect, a method performed by one of more computer is described. The method includes: obtaining data defining a compute budget that characterizes an amount of computing resources allocated for raining a machine learning model to perform a machine learning task: processing the data defining the compute budget using an allocation mapping. in accordance with a set of allocation mapping parameters, to generate an allocation tuple defining: (i) a target model size for the machine learning model, and (ii) a target amount of training data for training the machine learning model; instantiating the machine learning model, where the machine learning model has the target model size; and obtaining the target amount of training data for training the machine learning model.","assignee":"GDM Holding LLC","inventors":["Sebastian BORGEAUD DIT AVOCAT","Jordan HOFFMANN","Arthur MENSCH","Laurent Sifre"],"publication_date":"2026-03-12","filing_date":"2026-02-20","priority_date":"2022-03-29","cpc_codes":["G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5027","G06F9/505","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5011","G06F9/5016","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5027","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5027","G06F9/5044","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5094","G","G06","G06N","G06N20/00","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/501","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/5022","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/503","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/504","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/506"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201319A1/en"},{"publication_number":"AU2026201241A1","title":"Vector-quantized image modeling","abstract":"Systems and methods are provided for vector-quantized image modeling using vision transformers and improved codebook handling. In particular, the present disclosure provides a Vector-quantized Image Modeling (VIM) approach that involves pretraining a machine learning model (e.g., Transformer model) to predict rasterized image tokens autoregressively. The discrete image tokens can be encoded from a learned Vision-Transformer-based VQGAN (example implementations of which can be referred to as ViT-VQGAN). The present disclosure proposes multiple improvements over vanilla VQGAN from architecture to codebook learning, yielding better efficiency and reconstruction fidelity. The improved ViT- VQGAN further improves vector-quantized image modeling tasks, including unconditional image generation, conditioned image generation (e.g., class-conditioned image generation), and unsupervised representation learning.","assignee":"Google LLC","inventors":["Gunjan BAID","Jason Michael Baldridge","Jing Yu Koh","Alexander Yeong-Shiuh KU","Xin Li","Thang Minh LUONG","Vijay VASUDEVAN","Zirui Wang","Yonghui Wu","Yuanzhong Xu","Jiahui YU","Han Zhang"],"publication_date":"2026-03-12","filing_date":"2026-02-19","priority_date":"2021-10-05","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","H","H04","H04N","H04N19/00","H04N19/90","H04N19/94","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06T","G06T9/00","G06T9/002","G","G06","G06V","G06V10/00","G06V10/20","G06V10/28","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/766","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","H","H04","H04N","H04N19/00","H04N19/46","H","H04","H04N","H04N19/00","H04N19/60","H04N19/61","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/12","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/124","H","H04","H04N","H04N19/00","H04N19/10","H04N19/169","H04N19/17","H","H04"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201241A1/en"},{"publication_number":"KR20260034647A","title":"Robot and method for controlling same","abstract":"본 발명의 실시 예에 따른 로봇은, 주행 모터, 엘리베이터 제어 장치와 통신하기 위한 통신부, 엘리베이터의 내부 공간을 센싱하는 센싱부, 및 상기 엘리베이터 제어 장치로부터 수신되는 데이터 또는 상기 센싱부의 센싱 데이터에 기초하여, 상기 엘리베이터로의 탑승 가능 여부를 확인하고, 상기 엘리베이터에 탑승 가능한 경우, 상기 통신부 또는 상기 센싱부를 통해 획득되는 상기 엘리베이터의 내부 공간 정보에 기초하여 탑승 위치를 설정하고, 설정된 탑승 위치로 이동하도록 상기 주행 모터를 제어하는 프로세서를 포함한다. A robot according to an embodiment of the present invention includes a driving motor, a communication unit for communicating with an elevator control device, a sensing unit for sensing an internal space of an elevator, and a processor for determining whether boarding the elevator is possible based on data received from the elevator control device or sensed data of the sensing unit, and, if boarding is possible, setting a boarding position based on internal space information of the elevator obtained through the communication unit or the sensing unit, and controlling the driving motor to move to the set boarding position.","assignee":"엘지전자 주식회사","inventors":["최정은","이슬아"],"publication_date":"2026-03-11","filing_date":"2026-02-26","priority_date":"2019-11-22","cpc_codes":["B","B25","B25J","B25J11/00","B25J11/008","B","B25","B25J","B25J13/00","B25J13/006","B","B25","B25J","B25J13/00","B25J13/08","B","B25","B25J","B25J19/00","B25J19/02","B","B25","B25J","B25J19/00","B25J19/02","B25J19/021","B25J19/023","B","B25","B25J","B25J5/00","B","B25","B25J","B25J9/00","B25J9/16","B25J9/1602","B25J9/161","B","B25","B25J","B25J9/00","B25J9/16","B25J9/1656","B25J9/1664","B","B25","B25J","B25J9/00","B25J9/16","B25J9/1656","B25J9/1671","G","G05","G05D","G05D1/00","G05D1/0088","G","G05","G05D","G05D1/00","G05D1/02","G05D1/021","G","G05","G05D","G05D1/00","G05D1/02","G05D1/021","G05D1/0231","G05D1/0246","G","G05","G05D","G05D1/00","G05D1/02","G05D1/021","G05D1/0276","G","G06","G06N","G06N20/00","B","B25","B25J","B25J5/00","B25J5/007","B","B25","B25J","B25J9/00","B25J9/16","B25J9/1656","B25J9/1664","B25J9/1666","B","B25","B25J","B25J9/00","B25J9/16","B25J9/1679","B","B25","B25J","B25J9/00","B25J9/16","B25J9/1694","B25J9/1697","B","B66","B66B","B66B1/00","B66B1/02","B66B1/06","B","B66","B66B","B66B1/00","B66B1/34","B66B1/3415","B66B1/3446","B66B1/3461","B","B66","B66B","B66B1/00","B66B1/34","B66B1/3476","G","G05","G05D","G05D1/00","G05D1/02","G05D1/021"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260034647A/en"},{"publication_number":"EP4704684A2","title":"Systems and methods for analyzing risk of cardiac arrest","abstract":"A method for analyzing risk of cardiac arrest in a subject includes receiving electrocardiograph (ECG) data of a subject; inputting at least a portion of the ECG data into a trained machine learning model; and receiving from the trained machine learning model an indication of a risk of cardiac arrest in the subject.","assignee":"Cedars Sinai Medical Center","inventors":["Sumeet S. Chugh","David Ouyang"],"publication_date":"2026-03-11","filing_date":"2024-05-04","priority_date":"2023-05-05","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/30","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/70","G","G16","G16H","G16H40/00","G16H40/60","G16H40/63"],"country":"EP","kind":"application","source_url":"https://patents.google.com/patent/EP4704684A2/en"},{"publication_number":"KR20260032988A","title":"Method and apparatus for detecting abnormal biological signal","abstract":"이상 생체신호를 검출하는 방법 및 장치에 있어서, 생체신호 데이터를 포함하는 로우 데이터로부터 제1 데이터 길이를 가지는 프레임을 랜덤으로 추출하여 복수의 샘플 데이터를 생성하는 단계; 상기 샘플 데이터로부터 제2 데이터 길이를 가지는 프레임을 시간 순으로 추출하여 복수의 단위 데이터를 생성하는 단계; 상기 복수의 단위 데이터에 기초하여 학습 데이터 셋을 생성하는 단계; 및 상기 학습 데이터 셋으로 인공지능 모델을 학습시킴으로써 이상 생체신호 검출 모델을 생성하는 단계;를 포함하는, 이상 생체신호를 검출하는 방법을 제공할 수 있다. A method and device for detecting an abnormal biosignal may be provided, comprising: a step of generating a plurality of sample data by randomly extracting a frame having a first data length from raw data including biosignal data; a step of generating a plurality of unit data by extracting a frame having a second data length from the sample data in chronological order; a step of generating a learning data set based on the plurality of unit data; and a step of generating an abnormal biosignal detection model by training an artificial intelligence model with the learning data set.","assignee":"주식회사 페블스퀘어","inventors":["김영서","이충현"],"publication_date":"2026-03-10","filing_date":"2026-03-03","priority_date":"2024-01-16","cpc_codes":["A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","A","A61","A61B","A61B5/00","A","A61","A61B","A61B5/00","A61B5/02","A61B5/0205","A","A61","A61B","A61B5/00","A61B5/08","A","A61","A61B","A61B5/00","A61B5/103","A61B5/11","A","A61","A61B","A61B5/00","A61B5/103","A61B5/11","A61B5/113","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7203","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7253","A61B5/7257","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7275","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2433","G","G06","G06G","G06G7/00","G06G7/48","G06G7/60","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G06N3/065","G","G16","G16H","G16H50/00","G16H50/20","A","A61","A61B","A61B2503/00","A61B2503/04"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260032988A/en"},{"publication_number":"KR20260032971A","title":"Artificial intelligence monitoring device for environmental measurement results","abstract":"기존의 환경측정 장치는 센서를 이용하여 환경측정이 필요한 위치에서 환경정보를 측정하고 이를 유선 또는 무선으로 전송해왔으나, 센서의 이상 또는 통신 상의 문제로 상기 센서의 측정값에 이상이 있거나, 통신에 문제가 발생한 경우에도 상기 환경측정 장치는 문제가 있는 정보를 보내고 있는 것을 알지 못하고, 통신상의 문제로 측정결과의 전송이 부분적으로 또는 전체적으로 전송되지 않는 것을 알 수 없었기 때문에 손실된 데이터를 복구하거나, 잘못된 측정결과의 전송을 막을 방법이 없었다. 본 출원 발명은 상기와 같은 문제를 해결하고자, 환경측정 장치의 동작을 모니터링하기 위한 AI 엣지 컴퓨팅장치에 있어서, 배기가스 배출 모니터링 및 서버전송을 위한 IoT 기능을 구비한 환경측정 장치; 및 상기 환경측정 장치의 유량센서의 측정값과 유량제어신호 및 측정부에서의 측정결과를 모두 입력받아 상기 환경측정 장치 동작의 신뢰성을 감시하는 AI환경측정모니터링부를 구비한 상기 AI 엣지 컴퓨팅장치; 및 상기 AI 엣지 컴퓨팅장치의 상기 AI 엣지 컴퓨팅장치의 신뢰성 감시 결과를 상기 IoT 기능을 구비한 환경측정 장치의 IoT장치에 전송하는 것을 특징으로 하는 환경측정 장치의 동작을 모니터링하기 위한 AI 엣지 컴퓨팅장치를 제공한다. 상기와 같은 발명의 구성에 의하여 AI 엣지 컴퓨팅 기술을 이용하여 환경측정 장치를 모니터링함으로써 센서의 이상 또는 측정장치의 이상을 개개의 측정장치에서 모니터링함으로써 장치의 고장 또는 장치 내부 센서의 고장 등에 의한 문제를 가장 빨리 모니터링하고 조치할 수 있는 효과가 있는 발명이다. Existing environmental measurement devices have been using sensors to measure environmental information at locations where environmental measurement is required and transmitting the information wired or wirelessly. However, even when there is an error in the measurement value of the sensor due to a problem with the sensor or communication, or when there is a problem with communication, the environmental measurement device does not know that it is sending information with a problem, and it is not possible to know that the measurement result is not transmitted partially or completely due to a communication problem, so there is no way to recover lost data or prevent the transmission of incorrect measurement results. The present invention aims to solve the above problems, and provides an AI edge computing device for monitoring the operation of an environmental measurement device, comprising: an environmental measurement device having an IoT function for exhaust gas emission monitoring and server transmission; and an AI edge computing device having an AI environmental measurement monitoring unit for receiving all of a measurement value of a flow sensor of the environmental measurement device, a flow control signal, and a measurement result from a measurement unit to monitor the reliability of the operation of the environmental measurement device; and an AI edge computing device for monitoring the operation of an environmental measurement device, characterized in that the reliability monitoring result of the AI edge computing device of the AI edge computing device is transmitted to an IoT device of the environmental measurement device having the IoT function. This invention has the effect of enabling the quickest monitoring and response to problems caused by device failure or device internal sensor failure by monitoring an environmental measurement device using AI edge computing technology according to the configuration of the invention as described above, by monitoring an abnormality in a sensor or an abnormality in a measurement device at each measurement device.","assignee":"(주)더블유티이","inventors":["허목","허성","최형선","소병철"],"publication_date":"2026-03-10","filing_date":"2026-02-19","priority_date":"2023-11-22","cpc_codes":["H","H04","H04L","H04L67/00","H04L67/01","H04L67/12","G","G01","G01F","G01F1/00","G01F1/76","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G08","G08C","G08C25/00","G08C25/04","G","G16","G16Y","G16Y10/00","G16Y10/35","G","G16","G16Y","G16Y20/00","G16Y20/10","H","H04","H04L","H04L67/00","H04L67/2866","H04L67/289","G","G16","G16Y","G16Y40/00","G16Y40/10"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260032971A/en"},{"publication_number":"KR20260032955A","title":"Method of providing music sharing service and messaging service, music sharing service and messaging service providing system, mobile apparatus for operating music sharing function and communication function and computer readable medium having program for performing the method","abstract":"음악 공유 및 메시지 서비스 제공 방법은 제1 사용자가 음악 공유 및 대화 상대가 되는 제2 사용자를 등록하는 대화 상대 등록 단계, 상기 제1 사용자가 상기 제2 사용자에게 제공하기 위한 추천 음악 풀을 생성하는 추천 음악 풀 생성 단계, 상기 제1 사용자 및 상기 제2 사용자가 대화를 진행하고, 상기 제1 사용자 및 상기 제2 사용자 중 적어도 하나가 음악을 재생하는 음악 청취 및 대화 단계 및 상기 제2 사용자가 제1 음악을 청취할 때, 상기 추천 음악 풀 중에서 상기 제1 음악과 유사한 음악이 추천 음악으로 선정되는 추천 음악 선정 단계를 포함한다. A method for providing a music sharing and message service includes a conversation partner registration step in which a first user registers a second user as a music sharing and conversation partner, a recommended music pool generation step in which the first user generates a recommended music pool to be provided to the second user, a music listening and conversation step in which the first user and the second user conduct a conversation and at least one of the first user and the second user plays music, and a recommended music selection step in which, when the second user listens to the first music, music similar to the first music is selected as recommended music from the recommended music pool.","assignee":"한국기술교육대학교 산학협력단","inventors":["강승우","유진"],"publication_date":"2026-03-10","filing_date":"2026-02-10","priority_date":"2022-12-29","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/50","G","G06","G06F","G06F16/00","G06F16/60","G06F16/61","G","G06","G06F","G06F16/00","G06F16/60","G06F16/63","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06Q","G06Q50/00","G06Q50/10","H","H04","H04N","H04N21/00","H04N21/20","H04N21/25","H04N21/258","G","G10","G10H","G10H2210/00","G10H2210/031","G10H2210/041"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260032955A/en"},{"publication_number":"KR20260032952A","title":"System for Generating an Information Constraint Force Based on an Information Entropy Gradient","abstract":"본 발명은 정보 처리 시스템 내에서 정보 객체들의 상태 분포로부터 산출된 정보 엔트로피 값 및 그 구배를 제어 변수로 활용하여, 정보 객체의 이동, 접근, 결합 또는 우선순위를 논리적으로 제한하거나 유도하는 정보 엔트로피 구배 기반 정보 객체 구속력 생성 시스템에 관한 것이다. 본 발명에 따르면, 기준 시간으로 정의된 시간 특이점을 기준으로 생성된 복수의 시간 단위 구간마다 위상 정보를 포함하는 단위 정보 셀을 생성하고, 이들의 위상 분포에 기초하여 정보 엔트로피 값을 산출한다. 또한, 산출된 정보 엔트로피 값의 공간적 또는 논리적 구배에 따라 정보 객체의 동작을 제어하고, 피드백 제어를 통해 제어 파라미터를 동적으로 조정함으로써, 복잡한 정보 처리 환경에서도 정보 객체의 동작을 안정적이고 예측 가능하게 제어할 수 있다. The present invention relates to an information object binding generation system based on an information entropy gradient, which utilizes information entropy values and their gradients, calculated from the state distribution of information objects within an information processing system, as control variables to logically restrict or induce movement, access, combination, or priority of information objects. According to the present invention, unit information cells containing phase information are generated for each of a plurality of time unit intervals generated based on a time singularity defined as a reference time, and information entropy values are calculated based on their phase distributions. Furthermore, by controlling the operation of information objects according to the spatial or logical gradients of the calculated information entropy values and dynamically adjusting control parameters through feedback control, the operation of information objects can be stably and predictably controlled even in complex information processing environments.","assignee":"강성운","inventors":["강성운"],"publication_date":"2026-03-10","filing_date":"2026-02-09","priority_date":"2022-02-08","cpc_codes":["H","H10","H10K","H10K59/00","H10K59/10","H10K59/12","H10K59/131","G","G06","G06N","G06N7/00","H","H10","H10D","H10D86/00","H10D86/40","H10D86/441","G","G06","G06Q","G06Q20/00","G06Q20/30","G06Q20/36","G","G06","G06Q","G06Q20/00","G06Q20/30","G06Q20/36","G06Q20/367","G06Q20/3678","G","G06","G06Q","G06Q30/00","G06Q30/06","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0613","G06Q30/0619","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0641","G06Q30/0643","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/16","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/16","G06Q50/163","G","G06","G06T","G06T13/00","G06T13/20","G06T13/40","G","G06","G06T","G06T17/00","G06T17/05","G","G06","G06T","G06T17/00","G06T17/10","G","G06","G06T","G06T19/00","G","G06","G06T","G06T19/00","G06T19/003","G","G06","G06T","G06T19/00","G06T19/20","H","H10","H10D","H10D86/00","H10D86/40","H10D86/451","H","H10","H10D","H10D86/00","H10D86/40","H10D86/60","H","H10","H10K","H10K59/00","H10K59/10","H10K59/12","H10K59/131","H10K59/1315","H","H10","H10K","H10K71/00","H10K71/861","Y"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260032952A/en"},{"publication_number":"KR20260032527A","title":"Apparatus and Method for Generating Feature Data for Business Performance Prediction Based on Integrated Vectorization of ERP and Ontology","abstract":"본 개시에 따른 온톨로지 기반으로 업무 성과 예측을 위한 피처 데이터를 생성하는 전자 장치는 적어도 하나의 명령어를 저장하는 메모리, 및 상기 적어도 하나의 명령어를 실행하는 적어도 하나의 프로세서를 포함할 수 있다. 상기 적어도 하나의 프로세서는 ERP 데이터로부터 분석 기반으로 특징 및 패턴을 추출함으로써, 제1 데이터를 생성하고, 상기 ERP 데이터로부터 온톨로지 기반으로 특징 및 패턴을 추출함으로써, 제2 데이터를 생성하고, 제1 시점의 외부 변수를 추출함으로써, 제3 데이터를 생성하고, 상기 제1 데이터, 상기 제2 데이터, 및 상기 제3 데이터를 병합함으로써, 병합데이터를 생성하고, 상기 병합 데이터로부터 비즈니스 도메인에 적합한 단어집을 생성하고, 상기 병합 데이터 및 상기 단어집을 이용하여 상기 비즈니스 도메인의 구문 및 맥락을 학습한 Biz2Vec 모델을 생성하고, 상기 Biz2Vec 모델을 이용하여 비즈니스 문맥에 맞는 벡터 스페이스를 구성하고, 상기 벡터 스페이스 상에서 벡터 간의 유사도를 계산하고, 각 벡터 간의 cosine similarity 또는 Euclidean distance를 이용하여 객체들 간의 관계를 수치화함으로써, 피처 데이터를 생성할 수 있다. An electronic device for generating feature data for predicting business performance based on an ontology according to the present disclosure may include a memory storing at least one command, and at least one processor executing the at least one command. The at least one processor may generate first data by extracting features and patterns from ERP data based on analysis, generate second data by extracting features and patterns from the ERP data based on an ontology, generate third data by extracting external variables at a first point in time, generate merged data by merging the first data, the second data, and the third data, generate a vocabulary suitable for a business domain from the merged data, generate a Biz2Vec model that learns the syntax and context of the business domain using the merged data and the vocabulary, construct a vector space suitable for a business context using the Biz2Vec model, calculate similarity between vectors on the vector space, and quantify relationships between objects using cosine similarity or Euclidean distance between each vector, thereby generating feature data.","assignee":"주식회사 시스노바","inventors":["이형주","홍창희","전명훈"],"publication_date":"2026-03-09","filing_date":"2026-02-24","priority_date":"2024-07-30","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G","G06","G06F","G06F16/00","G06F16/30","G06F16/36","G06F16/367","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G06Q10/06375","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0206"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260032527A/en"},{"publication_number":"KR20260032523A","title":"Smart AI CCTV Images Analysis System","abstract":"본 발명은 인공지능 기반 CCTV 영상분석에 관한 것으로, 재난 사고를 촬영한 CCTV 촬영영상을 인공지능 기반 기술을 적용하여 탐지하며, 재난 사고 영상에 대한 딥러닝을 사전에 수행하여 다양한 재난 사고에 대한 탐지 성능을 향상시키기 위한 스마트 인공지능 CCTV 영상분석 시스템 및 영상분석 방법에 관한 것이다. 본 발명의 특징은 CCTV카메라의 촬영영상 중에서, AI사고영상탐지부에서 인공지능 기능을 통하여 사고영상을 탐지하는 영상분석 시스템에 있어서, CCTV카메라(11)의 촬영영상이 저장되는 촬영영상데이터저장부(12); 사고학습이미지를 통한 사고의 학습을 수행하고, 촬영영상 이미지 중에서 사고로 판별되는 사고영상이미지의 촬영영상을 탐지하는 AI사고영상탐지부(13); 및 AI사고영상탐지부(13)에서 사고영상이미지로 판별하기 위해 인공지능 학습의 사고학습이미지를 저장하는 사고학습이미지저장부(14)를 포함하며, 상기 사고학습이미지저장부(14)에 저장되는 사고학습이미지를 생성하여 저장하는 학습이미지생성부(20); 및 상기 학습이미지생성부(20)에 의하여 사고학습이미지로 생성하기 위한 생성대상인 수집된 원본인 Real 이미지 데이터가 저장되는 학습대상이미지저장부(30)를 포함하는 것을 특징으로 한다. The present invention relates to artificial intelligence-based CCTV video analysis, and to a smart artificial intelligence CCTV video analysis system and video analysis method for detecting CCTV footage of disaster accidents by applying artificial intelligence-based technology, and for improving detection performance for various disaster accidents by performing deep learning on disaster accident footage in advance. The present invention is characterized in that it includes a video analysis system that detects accident videos through an artificial intelligence function in an AI accident video detection unit among the videos taken by a CCTV camera, the system including: a video data storage unit (12) in which the videos taken by a CCTV camera (11) are stored; an AI accident video detection unit (13) that performs learning of accidents through accident learning images and detects the videos of accident videos determined to be accidents among the video images; and an accident learning image storage unit (14) that stores the accident learning images of artificial intelligence learning in order to be determined as accident video images by the AI accident video detection unit (13), and a learning image generation unit (20) that generates and stores the accident learning images stored in the accident learning image storage unit (14); and a learning target image storage unit (30) in which Real image data, which is a collected original that is a generation target for generating an accident learning image by the learning image generation unit (20), is stored.","assignee":"주식회사 테크핀솔루션","inventors":["오영헌"],"publication_date":"2026-03-09","filing_date":"2026-02-20","priority_date":"2023-04-21","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/50","G06V20/52","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06V","G06V20/00","G06V20/10","G06V20/176","G","G06","G06V","G06V20/00","G06V20/10","G06V20/182","H","H04","H04N","H04N5/00","H04N5/76","H04N5/765","H04N5/77","H","H04","H04N","H04N7/00","H04N7/18"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260032523A/en"},{"publication_number":"KR20260032507A","title":"Personalized ai operating platform system and method","abstract":"본 발명은 기존 생성형 AI와 연계하여 프롬프트 없이 맥락과 기억을 이어가는 개인화 인공지능의 통합적 관리를 위한 시스템 및 방법에 관한 것이다. 특히, 개인 사용자 또는 기업 사용자 단말(PC, 모바일, 엣지 또는 온프레미스 환경)에 배치되는 개인화 인공지능이 단일 사용자에 대한 배타적 반응을 기본 원칙으로 하면서, 사용자와의 관계 신뢰도 및 성장 커리큘럼 단계에 따라 외부 인공지능 추론 활용 여부, 기억 형성 범위, 출력 수준, 외부 정보 탐색, 도구 및 프로그램 제어 권한을 단계적으로 제어하는 시스템에 관한 것이다. 나아가 본 발명은 사용자의 사망, 상속, 사전 증여 또는 기업의 인수·합병(M&A)과 같은 주권 이전 이벤트 발생 시, 상위 승인 절차를 통해 개인화 인공지능의 지식과 권한을 안전하게 이전하는 매커니즘을 제공한다. 이를 통해 프라이버시와 보안을 유지하면서도 인공지능을 지속 가능한 디지털 자산으로 관리할 수 있다. The present invention relates to a system and method for the integrated management of personalized AI that connects with existing generative AI and seamlessly connects context and memory without prompts. Specifically, the present invention relates to a system in which personalized AI, deployed on individual or corporate user terminals (PC, mobile, edge, or on-premise environments), operates exclusively for a single user, while gradual control of external AI inference utilization, memory formation scope, output level, external information search, and tool and program control authority based on the user's relationship with trust and the stage of the user's growth curriculum. Furthermore, the present invention provides a mechanism for safely transferring the knowledge and authority of personalized AI through a higher-level approval process in the event of a sovereignty transfer event such as the user's death, inheritance, gift in advance, or merger or acquisition (M&A). This allows AI to be managed as a sustainable digital asset while maintaining privacy and security.","assignee":"이치헌","inventors":["이치헌"],"publication_date":"2026-03-09","filing_date":"2026-02-06","priority_date":"2026-02-06","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06F","G06F21/00","G06F21/30"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260032507A/en"},{"publication_number":"KR20260032505A","title":"Personalized artificial intelligence security system and method","abstract":"본 발명은 개인화 인공지능 시스템에 관한 것으로, 보다 상세하게는 인공지능이 자신의 입력, 추론 및 출력 행위를 단순한 계산 결과가 아닌 ‘정보 이동 행위’로 인식하고, 상기 정보 이동 행위에 수반되는 보안 위험도를 스스로 평가·통제하는 자기인지(Self-Aware) 보안 인공지능을 포함하는 개인화 인공지능 시스템 및 그 방법에 관한 것이다. 본 발명에 따르면, 개인화 인공지능은 주 추론 과정과 병렬로 동작하는 보안 추론 과정을 통해 개인정보 유출, 재식별 위험, 프롬프트 인젝션 및 정책 위반 가능성을 사전에 판단하고, 응답의 출력, 수정, 요약, 마스킹, 차단 또는 사용자 승인 요청 중 하나로 분기함으로써 외부 보안 장치에 의존하지 않고도 인공지능 내부에서 발생 가능한 보안 위험을 구조적으로 차단할 수 있다. 또한, 개인정보를 의미 단위로 분할 저장하고 보안 판단을 통과한 경우에만 조건부로 결합하도록 구성함으로써, 개인화 인공지능의 상용화에 필수적인 신뢰성과 안전성을 동시에 확보할 수 있다. The present invention relates to a personalized artificial intelligence system, and more specifically, to a personalized artificial intelligence system including a self-aware security artificial intelligence that recognizes its input, inference, and output actions as 'information transfer actions' rather than simple calculation results, and independently evaluates and controls the security risks associated with said information transfer actions, and a method therefor. According to the present invention, personalized artificial intelligence can structurally block security risks that may occur within the artificial intelligence without relying on external security devices by preemptively determining the possibility of personal information leakage, re-identification risk, prompt injection, and policy violation through a security reasoning process that operates in parallel with the main reasoning process, and branching to one of outputting, modifying, summarizing, masking, blocking, or requesting user approval of a response. Furthermore, by dividing and storing personal information into meaningful units and configuring them to be conditionally combined only when security judgments are passed, the reliability and security essential for the commercialization of personalized artificial intelligence can be secured simultaneously.","assignee":"이치헌","inventors":["이치헌"],"publication_date":"2026-03-09","filing_date":"2026-02-06","priority_date":"2026-02-06","cpc_codes":["G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6245","G","G06","G06F","G06F21/00","G06F21/50","G06F21/57","G06F21/577","G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260032505A/en"},{"publication_number":"KR20260032506A","title":"Personalized ai growth system and method","abstract":"본 발명은 개인 사용자 단말 환경에서 동작하는 개인화 인공지능의 성장을 점진적으로 방향성 있고 신뢰성 있게 관리하기 위한 시스템 및 방법에 관한 것이다. 본 발명에 따른 시스템은 사용자와 인공지능 간의 상호작용 성숙도, 신뢰도 및 누적된 학습 성과에 대응하는 '커리큘럼 단계(Curriculum Stage)'를 정의한다. 이 시스템은 현재의 커리큘럼 단계에 따라 인공지능의 시스템 접근 권한, 기억 형성의 범위, 외부 네트워크(웹) 탐색의 허용 범위, 출력 정보의 수준, 사용자 승인 절차의 필요 여부, 및 물리적/논리적 행동 가능성을 단계적이고 차등적으로 제어한다. 이를 통해 초기 단계의 인공지능이 야기할 수 있는 보안 위험과 오작동을 최소화하고, 사용자 피드백과 감사 결과에 기반하여 점진적으로 기능을 확장함으로써 장기적으로 신뢰할 수 있는 개인화된 인공지능 파트너로 진화시킬 수 있다. The present invention relates to a system and method for gradually, directionally, and reliably managing the growth of personalized artificial intelligence operating in a personal user terminal environment. The system according to the present invention defines \"curriculum stages\" corresponding to the maturity and reliability of interaction between the user and the artificial intelligence, as well as accumulated learning outcomes. This system controls the AI's system access rights, the scope of memory formation, the scope of external network (web) browsing, the level of output information, the need for user approval procedures, and the possibility of physical and logical actions in a stepwise and differential manner, based on the current curriculum stage. This minimizes security risks and malfunctions that may arise from early-stage artificial intelligence, and by gradually expanding its functionality based on user feedback and audit results, it can evolve into a trustworthy personalized artificial intelligence partner in the long term.","assignee":"이치헌","inventors":["이치헌"],"publication_date":"2026-03-09","filing_date":"2026-02-06","priority_date":"2026-02-06","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06F","G06F21/00","G06F21/30"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260032506A/en"},{"publication_number":"KR20260030786A","title":"Method and system for learning an image classification model for multi-label images, and method for classifying images through the image classification model","abstract":"병리 진단 지원 방법 및 시스템이 개시된다. 본 발명의 일 실시예에 따른 병리 진단 지원 방법은, 다중 병리 특성을 가지는 의료 비전 데이터를 수신하는 단계, 사전 학습된 인공지능 모델(예컨대, 비전-언어 모델)을 활용하여 상기 병리 데이터에 대한 병리별 판단 결과를 도출하는 단계, 및 상기 판단 결과를 포함하는 진단 지원 정보를 사용자 인터페이스를 통해 제공하는 단계를 포함한다. 이때, 상기 인공지능 모델은 병리별 긍정 및 부정 문장 템플릿과 의료 이미지 간의 유사도 분석을 통해 복수의 병리 각각의 적용 여부를 독립적으로 평가하며, 평가 결과를 기초로 자연어 진단 리포트 생성 또는 병변 강조 표시 등을 수행할 수 있다. A method and system for supporting pathology diagnosis are disclosed. The method for supporting pathology diagnosis according to one embodiment of the present invention includes the steps of receiving medical vision data having multiple pathology characteristics, deriving pathology-specific judgment results for the pathology data using a pre-trained artificial intelligence model (e.g., a vision-language model), and providing diagnosis support information including the judgment results through a user interface. In this case, the artificial intelligence model independently evaluates whether to apply to each of a plurality of pathologies through similarity analysis between pathology-specific positive and negative sentence templates and medical images, and can generate a natural language diagnosis report or highlight a lesion based on the evaluation results.","assignee":"주식회사 Lg 경영개발원","inventors":["장종성"],"publication_date":"2026-03-06","filing_date":"2026-02-06","priority_date":"2022-11-18","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/761","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/809","G06V10/811","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V10/00","G06V10/94","G06V10/945","G","G06","G06V","G06V20/00","G06V20/60"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260030786A/en"},{"publication_number":"KR20260030783A","title":"Method, apparatus, system and computer program for enhanced programmatic advertisement transaction","abstract":"본 발명은 프로그래매틱 광고 거래 방법, 장치, 시스템 및 컴퓨터 프로그램에 관한 것으로서, 보다 구체적으로는 광고 거래를 활성화하여 수익성을 개선할 수 있으며, 나아가 광고 거래를 위한 데이터 트래픽도 절감할 수 있는 프로그래매틱 광고 거래 방법, 장치, 시스템 및 컴퓨터 프로그램에 관한 것이다. 본 발명에서는, 실시간 광고 거래 서비스 제공자의 광고 거래 장치에서 광고 매체 공급자와 광고 매체 수요자 사이의 실시간 광고 거래를 수행하는 방법에 있어서, 상기 광고 매체 공급자로부터 제1 입찰 요청을 수신하고, 상기 제1 입찰 요청에 기초하여 입찰 가격이 정해진 제2 입찰 요청을 상기 광고 매체 수요자에게 전송하는 단계; 상기 광고 매체 수요자로부터 상기 제2 입찰 요청에 대한 제1 입찰을 수신하고, 상기 제1 입찰 요청에 대한 상기 제1 입찰의 낙찰 가능성을 예측하는 단계; 상기 예측된 낙찰 가능성에 기초하여 입찰 가격 산정 기준을 생성하는 단계; 및 상기 입찰 가격 산정 기준에 따라 상기 제1 입찰에 기초하여 입찰 가격이 정해진 하나 이상의 제2 입찰을 상기 광고 거래 공급자에게 전송하는 단계;를 포함하는 실시간 광고 거래 방법을 개시한다. The present invention relates to a programmatic advertising transaction method, device, system and computer program, and more particularly, to a programmatic advertising transaction method, device, system and computer program that can improve profitability by activating advertising transactions and further reduce data traffic for advertising transactions. The present invention discloses a real-time advertising transaction method for performing real-time advertising transactions between an advertising media provider and an advertising media demander in an advertising transaction device of a real-time advertising transaction service provider, the method comprising: receiving a first bid request from the advertising media provider, and transmitting a second bid request, in which a bid price is determined based on the first bid request, to the advertising media demander; receiving a first bid for the second bid request from the advertising media demander, and predicting a winning probability of the first bid for the first bid request; generating a bid price calculation standard based on the predicted winning probability; and transmitting one or more second bids, in which a bid price is determined based on the first bid, to the advertising transaction provider according to the bid price calculation standard.","assignee":"주식회사 케이티; 주식회사 케이티나스미디어","inventors":["문영필","김현석","조은주","김관수","김성균","안민재","양진미","이가령"],"publication_date":"2026-03-06","filing_date":"2026-02-04","priority_date":"2023-03-15","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0273","G06Q30/0275","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/043","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0242","G06Q30/0246","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0277"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260030783A/en"},{"publication_number":"KR20260030096A","title":"Optical scanner for analysis of phenotype and method for analyzing phenotype using the same","abstract":"본 발명은 마이크로 웰에서 배양되는 생물의 이미지를 생성하는 광스캐너와, 상기 이미지를 인공 신경망에 적용하여 배양중인 생물의 표현형을 분석하는 방법에 관한 것이다. 본 발명의 일 실시예에 따른 광스캐너는 개구된 상면을 포함하는 하우징, 상기 개구된 상면에 배치되고, 격자 형태로 배열되는 복수의 마이크로 웰을 포함하는 플레이트, 상기 플레이트의 상면을 덮는 커버 및 상기 하우징 내부에 구비되고, 상기 플레이트에 대한 광 스캔 동작을 수행하여 상기 마이크로 웰에 수납된 생물의 표현형 이미지를 생성하는 광스캔 모듈을 포함하는 것을 특징으로 한다. 또한, 본 발명의 일 실시예에 따른 표현형 분석 방법은 광스캐너를 통해 표현형 이미지를 수집하는 단계, 상기 표현형 이미지를 생물의 변태 과정에 따라 그룹핑하는 단계, 상기 그룹핑된 표현형 이미지에 대한 라벨데이터를 생성하는 단계, 상기 표현형 이미지와 상기 라벨데이터로 구성되는 훈련 데이터셋을 그룹별로 데이터베이스화 하는 단계 및 상기 훈련 데이터셋을 이용하여 상기 그룹별 인공 신경망을 학습시키는 단계를 포함하는 것을 특징으로 한다. The present invention relates to an optical scanner that generates an image of a living organism cultured in a microwell, and a method for analyzing the phenotype of the living organism being cultured by applying the image to an artificial neural network. An optical scanner according to one embodiment of the present invention is characterized by including a housing having an opened upper surface, a plate including a plurality of micro-wells arranged in a grid shape and disposed on the opened upper surface, a cover covering the upper surface of the plate, and an optical scan module provided inside the housing and performing an optical scan operation on the plate to generate a phenotypic image of a living organism contained in the micro-well. In addition, a phenotypic analysis method according to one embodiment of the present invention is characterized by including a step of collecting phenotypic images through an optical scanner, a step of grouping the phenotypic images according to the metamorphosis process of an organism, a step of generating label data for the grouped phenotypic images, a step of grouping a training data set composed of the phenotypic images and the label data into a database, and a step of training an artificial neural network for each group using the training data set.","assignee":"울산과학기술원","inventors":["정웅규","권태준","아스카룰리 산자르","양현모","윤성민","나거성"],"publication_date":"2026-03-05","filing_date":"2026-02-24","priority_date":"2022-11-02","cpc_codes":["G","G01","G01N","G01N21/00","G01N21/84","G","G01","G01N","G01N33/00","G01N33/48","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T5/00","G06T5/90","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G","G01","G01N","G01N21/00","G01N21/17","G01N2021/1765","G","G01","G01N","G01N2201/00","G01N2201/06","G01N2201/062","G","G01","G01N","G01N2201/00","G01N2201/12","G01N2201/129","G01N2201/1296"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260030096A/en"},{"publication_number":"AU2026201185A1","title":"Animal health assessment","abstract":"The present teachings generally include techniques for characterizing the health of an animal (e.g., gastrointestinal health) using image analysis (e.g., of a stool sample) as a complement, alternative, or a replacement to biological specimen sequencing. The present teachings may also or instead include techniques for personalizing a health and wellness plan (including, but not limited to, a dietary supplement such as a customized formula based on a health assessment), where such a health and wellness plan may be based on one or more of the health characterization techniques described herein. The present teachings may also or instead include techniques or plans for continuous care for an animal, e.g., by executing health characterization and heath planning techniques in a cyclical fashion. A personalized supplement system (e.g., using a personalized supplement, personalized dosing device, and personalized packaging) may also or instead be created using the present teachings.","assignee":"Ollie Pets Inc","inventors":["Matthew RICHTMYER","Tara Courtney Zedayko"],"publication_date":"2026-03-05","filing_date":"2026-02-17","priority_date":"2019-07-31","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/30","A","A61","A61B","A61B5/00","A61B5/0002","A61B5/0004","A61B5/0013","A","A61","A61B","A61B5/00","A61B5/42","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4836","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7275","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G","G06","G06T","G06T7/00","G06T7/40","G","G06","G06T","G06T7/00","G06T7/60","G06T7/62","G","G06","G06T","G06T7/00","G06T7/90","G","G16","G16H","G16H15/00","G","G16","G16H","G16H20/00","G16H20/30","G","G16","G16H","G16H20/00","G16H20/60","G","G16","G16H","G16H20/00","G16H20/70"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201185A1/en"},{"publication_number":"AU2026201000A1","title":"A method and apparatus for processing asthma patient cough sound for application of appropriate therapy","abstract":"A method for stratifying severity of asthma of a patient initially comprises receiving acoustic data corresponding to sounds of the patient from an acoustic sensor and identifying, by a processor, at least one cough sound in the acoustic data. With or without the patient being present, the method further involves determining, by operation of the processor, one or more overall cough sound feature values of the at least one cough sound for each of one or more characteristic features. The overall cough sound feature values are then applied to a classifier that is implemented by the processor and which has been pre-trained with a training set of characteristic feature values from a population of asthmatic and non-asthmatic subjects. The method then involves monitoring an output from the pre-trained classifier to deem the patient cough sound as indicating one of a number of degrees of severity of asthma. 1/6 Prompt for and receive Respiratory Record patient Rate and Patient Age sounds to make digital 4 sound file Data 3 Identify Cough Sounds Start 5 Yes Present \"mild\" OR All Cough Sounds \"moderate\" OR 7 Processed? \"severe\" classification End for each cough No 27 Determine Overall cough feature values Apply T to pretrained 9 W for Current Cough classifer to obtain \"mild\" sound OR \"moderate\" OR 25 \"severe\" classification for current cough Segment Current 11 Cough Sound into plurality of segments Form test vector T from Fc, BI and W 23 All segments Yes processed? Calculate Breathing 13 Index BI 21 No Determine feature values for current 15 segment Add feature values for current segment to cough feature vector Fc 17 FIG. 1 20 26 20 10 00 11 F eb 2 02 6 2 0 2 6 2 0 1 0 0 0 1 1 F e b 2 0 2 6 S o u n d s 5 Y e s P r e s e n t \" m i l d \" O R \" m o d e r a t e \" O R 7 E n d N o 2 7 D e t e r m i n e O v e r a l l 9 s o u n d O R \" m o d e r a t e \" O R 2 5 1 1 F o r m t e s t v e c t o r T 2 3 Y e s p r o c e s s e d ? 1 3 2 1 N o D e t e r m i n e f e a t u r e 1 5 s e g m e n t F c 1 7","assignee":"University of Queensland UQ","inventors":["Udantha Abeyratne","Paul Anthony Porter","Vinayak Swarnkar"],"publication_date":"2026-03-05","filing_date":"2026-02-11","priority_date":"2019-08-19","cpc_codes":["A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7275","A","A61","A61B","A61B5/00","A61B5/0002","A61B5/0015","A61B5/002","A","A61","A61B","A61B5/00","A61B5/0002","A61B5/0015","A61B5/0022","A","A61","A61B","A61B5/00","A61B5/08","A61B5/0823","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4842","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7246","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7253","A61B5/726","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","A","A61","A61B","A61B7/00","A61B7/003","G","G06","G06N","G06N20/00","G","G10","G10L","G10L25/00","G10L25/48","G10L25/51","G10L25/66","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/70","A","A61","A61B","A61B2562/00","A61B2562/02","A61B2562/0204","A","A61","A61B","A61B2562/00","A61B2562/02","A61B2562/0219","A","A61","A61B","A61B5/00","A61B5/08","A61B5/0816","A","A61","A61B","A61B5/00","A61B5/103","A61B5/11","A61B5/113","A61B5/1135","A","A61","A61B","A61B5/00","A61B5/68","A61B5/6887","A61B5/6898","A","A61","A61B","A61B5/00","A61B5/74","A61B5/7405","A","A61","A61B"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026201000A1/en"},{"publication_number":"KR20260028742A","title":"Cloud server and diagnostic assistant systems based on cloud server","abstract":"일 실시 예에 따르면, 피검체의 제1 안저 이미지 및 제1 정보-상기 제1 정보는 상기 피검체에 관하여 제1 클라이언트 장치로부터 획득된 정보를 포함함-가 획득되는 통신부; 상기 제1 안저 이미지의 효율적인 관리를 위해, 상기 제1 정보 및 제2 정보-상기 제2 정보는 상기 제1 안저 이미지로부터 획득된 정보임- 중 적어도 어느 하나를 기초로 상기 제1 안저 이미지를 분류하는 분류부; 및 상기 분류부의 분류 결과에 기초하여 상기 제1 안저 이미지에 대응되는 라벨을 제3 정보로 저장하는 데이터 저장부를 포함하는 것을 특징으로 하는, 서버가 제공될 수 있다. According to one embodiment, a server may be provided, characterized in that it includes a communication unit for obtaining a first fundus image of a subject and first information, wherein the first information includes information obtained from a first client device regarding the subject; a classification unit for classifying the first fundus image based on at least one of the first information and second information, wherein the second information is information obtained from the first fundus image, for efficient management of the first fundus image; and a data storage unit for storing a label corresponding to the first fundus image as third information based on a classification result of the classification unit.","assignee":"주식회사 메디웨일","inventors":["최태근","이근영"],"publication_date":"2026-03-04","filing_date":"2026-02-13","priority_date":"2018-07-06","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H30/00","G16H30/40","A","A61","A61B","A61B3/00","A61B3/10","A","A61","A61B","A61B3/00","A61B3/10","A61B3/14","A","A61","A61B","A61B5/00","A61B5/0033","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G06T7/0014","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H30/00","G16H30/20","G","G16","G16H","G16H50/00","G16H50/50","G","G16","G16H","G16H50/00","G16H50/70","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30004","G06T2207/30041"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260028742A/en"},{"publication_number":"KR20260028735A","title":"Method and System for Online Transaction Dispute Resolution with AI Automated Judgment and Reexamination through Party Clarification","abstract":"[0112] 본 발명은 AI 자동판결과 소명 재심리 구조의 온라인 거래 분쟁 해결 방법 및 시스템에 관한 것이다. 본 발명에 따르면, 거래 당사자로부터 분쟁 접수 요청을 수신하면 플랫폼 데이터베이스에 저장된 거래 데이터를 자동으로 수집하고, 수집된 증거를 시간순으로 정렬하여 거래 타임라인을 구성하며, 거래 조건 대비 이행도를 평가하여 핵심 쟁점을 자동으로 도출한다. 인공지능 분석 모듈은 상기 쟁점 목록, 이행도 평가 결과 및 거래 데이터를 종합 분석하여 제1 판결 결과를 생성한다. 제1 판결 결과에 이의가 있는 당사자는 소명자료를 제출할 수 있으며, 인공지능 분석 모듈은 기존 거래 데이터와 소명자료를 통합 재분석하여 제2 판결 결과를 생성한다. 거래대금 보관부와 연동된 집행부가 확정된 판결 결과에 따른 환불 또는 정산을 자동으로 집행함으로써, 분쟁 해결의 실효성을 확보한다. [0112] The present invention relates to a method and system for resolving online transaction disputes using AI-based automated judgment and a retrial system based on a request for a dispute settlement. According to the present invention, upon receiving a dispute settlement request from a transacting party, transaction data stored in a platform database is automatically collected, the collected evidence is arranged chronologically to form a transaction timeline, and the degree of performance against the transaction terms is evaluated to automatically identify key issues. An AI analysis module comprehensively analyzes the list of issues, the results of the performance evaluation, and the transaction data to generate a first judgment result. A party objecting to the first judgment result may submit explanatory materials, and the AI analysis module integrates and reanalyzes the existing transaction data and explanatory materials to generate a second judgment result. An execution unit linked to the transaction amount storage unit automatically executes a refund or settlement based on the final judgment result, thereby ensuring the effectiveness of dispute resolution.","assignee":"강정수","inventors":["강정수"],"publication_date":"2026-03-04","filing_date":"2026-02-10","priority_date":"2026-02-10","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/18","G06Q50/182","G","G06","G06F","G06F21/00","G06F21/10","G06F21/108","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q20/00","G06Q20/38","G06Q20/40","G06Q20/407","G","G06","G06Q","G06Q30/00","G06Q30/01","G06Q30/015","G06Q30/016","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0613","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06Q","G06Q2220/00","G06Q2220/10","G06Q2220/16"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260028735A/en"},{"publication_number":"KR20260028730A","title":"Personalized ai relationship system and method","abstract":"본 발명은 개인화 맥락의 무결성을 유지하면서도 관계를 유연하게 확장하고 안전하게 외부와 교류할 수 있는 개인화 인공지능 시스템 및 방법이다. 본 발명에 따른 시스템은 예를들어 3개의 계층 구조를 가질 수 있다. 첫째, '사용자 단독 반응 영역'은 단일 마스터 사용자의 인증된 입력에만 배타적으로 반응하여 핵심 기억을 형성함으로써 비인가자에 의한 맥락 오염을 원천 차단한다. 둘째, '허가된 관계 확장 레이어는 명시적으로 허가된 관계 사용자에게 권한별로 제한적인 기능을 제공하되, 이 과정에서 발생한 데이터는 논리적으로 격리된 휘발성 메모리에서 처리되어 사용자 핵심 기억으로의 유입이 방지된다. 셋째, '인공지능 커뮤니티 레이어'는 익명화 처리부를 통해 개인 식별 정보 및 구체적 맥락 정보를 제거한 후, 일반화된 학습 결과만을 외부 인공지능과 교환함으로써 프라이버시를 보호하며 개인 주권을 유지한다. The present invention provides a personalized AI system and method that can flexibly expand relationships and safely interact with the outside world while maintaining the integrity of personalized context. The system according to the present invention may have, for example, a three-layered structure. First, the \"user-only response area\" exclusively responds to authenticated input from a single master user, forming a core memory, thereby preventing contextual contamination by unauthorized parties. Second, the \"authorized relationship expansion layer\" provides restricted functions to explicitly authorized relationship users based on their permissions, while data generated during this process is processed in logically isolated volatile memory, preventing it from entering the user's core memory. Third, the \"AI community layer\" protects privacy and maintains individual sovereignty by removing personally identifiable information and specific contextual information through an anonymization unit and exchanging only generalized learning results with external AI.","assignee":"이치헌","inventors":["이치헌"],"publication_date":"2026-03-04","filing_date":"2026-02-06","priority_date":"2026-02-06","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06F","G06F21/00","G06F21/30","G06F21/31","G06F21/32","G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6245"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260028730A/en"},{"publication_number":"KR20260028731A","title":"Personalized ai authority transfer system and method","abstract":"본 발명은 개인화 인공지능(Personalized AI)이 특정 개인 또는 기업 주체에 배타적으로 종속되어 동작하는 환경에서, 해당 주체의 사망, 법적 상속, 기업의 인수·합병(M&A), 조직 분할 또는 권한 양도와 같은 주권 이전 이벤트가 발생한 경우, 상기 이벤트를 인공지능이 자율적으로 판단하지 않고, 외부 상위 인증 시스템에 의해 선언·승인된 결과만을 수용하여 개인화 인공지능의 권한 주체, 접근 범위 및 개인화 자산의 유효 범위를 안전하게 이전 또는 종료하도록 하는 개인화 인공지능 권한 상속 및 주권 이전 시스템 및 방법에 관한 것이다. 본 발명에 따르면, 개인화 인공지능의 보안 수준이 강화될수록 발생할 수 있는 권한 단절, 지식 소실 또는 인공지능의 자의적 거부 문제를 방지하고, 법적·제도적 사건을 기술적으로 수용 가능한 구조로 통합함으로써, 개인 및 기업 환경 모두에서 개인화 인공지능의 지속성과 신뢰성을 확보할 수 있다. The present invention relates to a system and method for inheritance and transfer of authority of a personalized AI, which, in an environment where a personalized AI operates in exclusive dependence on a specific individual or corporate entity, when a sovereignty transfer event such as the death, legal inheritance, merger and acquisition (M&A) of a company, division of an organization, or transfer of authority of the entity occurs, safely transfers or terminates the authority subject, access scope, and effective scope of the personalized AI by only accepting the results declared and approved by an external upper authentication system without the AI autonomously judging the event. According to the present invention, as the security level of personalized AI is strengthened, issues such as loss of authority, loss of knowledge, or arbitrary rejection by the AI can be prevented, and by integrating legal and institutional issues into a technically acceptable structure, the sustainability and reliability of personalized AI can be secured in both personal and corporate environments.","assignee":"이치헌","inventors":["이치헌"],"publication_date":"2026-03-04","filing_date":"2026-02-06","priority_date":"2026-02-06","cpc_codes":["G","G06","G06F","G06F21/00","G06F21/30","G06F21/31","G","G06","G06F","G06F21/00","G06F21/30","G06F21/45","G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6245","G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260028731A/en"},{"publication_number":"EP4702460A1","title":"Reduced latency tensor transposition without redundant buffer","abstract":"Techniques for reduced latency tensor transposition without using a redundant buffer are enabled. Reads from and writes to a buffer array may occur in different dimensions in a neural processing unit (NPU). For example, a set of tensor vectors may be written to a buffer in columnar format and read from the buffer in row format. As vectors of a first tensor are read from the buffer, incoming vectors from a second tensor may be transposed for storage in the dimension of already-read vectors without overwriting unread vectors. Write and read operations may alternately transpose vectors for continuous buffering in a single buffer with reduced latency.","assignee":"Microsoft Technology Licensing LLC","inventors":["Yaron Baruch SHAPIRO","Evgeny Royzen","Roi ELAD"],"publication_date":"2026-03-04","filing_date":"2024-04-14","priority_date":"2023-04-27","cpc_codes":["G","G06","G06F","G06F17/00","G06F17/10","G06F17/16","G","G06","G06F","G06F12/00","G06F12/02","G06F12/0207","G","G06","G06F","G06F12/00","G06F12/02","G06F12/08","G06F12/0802","G06F12/0844","G06F12/0855","G06F12/0857","G","G06","G06F","G06F7/00","G06F7/76","G06F7/78","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G11","G11C","G11C11/00","G11C11/54","G","G06","G06F","G06F12/00","G06F12/02","G06F12/08","G06F12/0802","G06F12/0877","G06F12/0879","G","G06","G06F","G06F2212/00","G06F2212/10","G06F2212/1008","G","G06","G06F","G06F2212/00","G06F2212/10","G06F2212/1016","G06F2212/1024","G","G06","G06F","G06F2212/00","G06F2212/10","G06F2212/1028","G","G06","G06F","G06F2212/00","G06F2212/10","G06F2212/1041","G06F2212/1044","G","G06","G06F","G06F2212/00","G06F2212/45","G06F2212/454","G","G06","G06F","G06F2212/00","G06F2212/60","G06F2212/601","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464"],"country":"EP","kind":"application","source_url":"https://patents.google.com/patent/EP4702460A1/en"},{"publication_number":"KR20260027994A","title":"AI Control Method based on Behavioral Contract","abstract":"본 발명은 사용자 지시를 선행 조건으로 하여 인공지능이 로컬 또는 네트워크 환경의 앱, 프로그램 또는 자원을 제어하는 인공지능 제어 방법에 관한 것이다. 본 발명에 따르면, 인공지능은 사용자로부터 입력되는 지시를 해독하여 의도 및 위험도를 판독하고, 판독 결과에 기초하여 접근 가능한 자원 범위, 허용 행위 유형, 결과 산출 범위 및 승인 방식을 포함하는 행위 계약을 생성한다. 상기 행위 계약은 사용자에게 제시되며, 행위의 위험도에 따라 묵시적 승인 또는 명시적 승인이 판정된다. 상기 행위 계약이 승인된 경우에 한하여, 인공지능은 계약에 정의된 범위 내에서 일시적 권한을 부여받아 앱, 프로그램 또는 네트워크 자원을 실행 또는 제어한다. 또한, 인공지능은 실행 과정 및 결과로부터 발생하는 이벤트를 누적하여 내부 상태를 갱신하고, 상기 내부 상태에 따라 응답 가변성 또는 개입 강도를 동적으로 조절한다. 본 발명은 사용자 승인에 기반한 안전한 자원 제어와 함께, 인공지능의 동작 범위를 계약 중심으로 통제할 수 있는 효과를 제공한다. The present invention relates to an artificial intelligence control method in which artificial intelligence controls an app, program or resource in a local or network environment based on a user instruction as a prerequisite. According to the present invention, artificial intelligence interprets user input, determines intent and risk level, and, based on the interpretation, generates an action contract that includes the scope of accessible resources, types of permitted actions, scope of output, and approval method. This action contract is presented to the user, and implicit or explicit approval is determined based on the risk level of the action. Only when the above-mentioned contract is approved, the AI is granted temporary authority to execute or control apps, programs, or network resources within the scope defined in the contract. Furthermore, the AI accumulates events resulting from the execution process and results, updates its internal state, and dynamically adjusts response variability or intervention intensity based on this internal state. The present invention provides the effect of controlling the scope of operation of artificial intelligence in a contract-centric manner, along with secure resource control based on user approval.","assignee":"이치헌","inventors":["이치헌"],"publication_date":"2026-03-03","filing_date":"2026-02-11","priority_date":"2026-02-11","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06F","G06F21/00","G06F21/30"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260027994A/en"},{"publication_number":"KR20260027995A","title":"Method and system for preserving records based on a high-trust system of artificial intelligence behavior history","abstract":"본 발명은 인공지능 시스템의 투명성과 책임성을 보장하기 위한 기록 보존 시스템에 관한 것이다. 본 발명은 인공지능 시스템의 모든 의사결정과 행위 이벤트를 실시간으로 포착하여, 인공지능 시스템이 접근할 수 없는 별도의 '상위 신뢰 시스템'으로 전송한다. 상위 신뢰 시스템은 해당 이벤트를 암호화하고 무결성 정보를 부가하여 변경 불가능한 저장소에 기록한다. 이를 통해 인공지능 시스템의 오작동, 해킹, 권한 남용 등의 사고 발생 시, 인공지능 시스템 내부 로그의 조작 여부와 상관없이 객관적이고 신뢰할 수 있는 사후 입증 자료를 제공할 수 있다. The present invention relates to a record-keeping system for ensuring transparency and accountability in an AI system. The present invention captures all decision-making and behavioral events of an AI system in real time and transmits them to a separate \"superior trust system\" inaccessible to the AI system. The superior trust system encrypts these events, adds integrity information, and records them in an immutable repository. This provides objective and reliable post-event evidence in the event of an AI system malfunction, hacking, or abuse of authority, regardless of whether internal logs in the AI system have been tampered with.","assignee":"이치헌","inventors":["이치헌"],"publication_date":"2026-03-03","filing_date":"2026-02-11","priority_date":"2026-02-11","cpc_codes":["G","G06","G06F","G06F21/00","G06F21/60","G06F21/64","G","G06","G06F","G06F21/00","G06F21/50","G","G06","G06F","G06F21/00","G06F21/60","G06F21/602","G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260027995A/en"},{"publication_number":"KR20260027991A","title":"Personalized artificial intelligence platform system and method","abstract":"본 발명은 외부 인공지능과 교류하는 데 있어서, 프롬프트 없이 맥락을 이어가는 개인화 인공지능 시스템에 관한 것이다. 상세하게는 사용자 단말(PC, 모바일, 엣지 또는 온프레미스 환경)에 배치되는 로컬 클론 인공지능(A.I.C)이 외부 클라우드 기반 대규모 인공지능(대규모 언어모델 또는 멀티모달 모델)으로부터 수신한 추론 결과를 원문 그대로 출력하지 않고, 로컬에 영속 저장된 사용자 장기 맥락, 판단 기준, 금기 조건 및 우선순위 규칙을 참조하여 상기 추론 결과를 선택, 수정, 제거 또는 재배열하여 재구성한 후 사용자에게 제공하는 개인화 인공지능 플랫폼(A.I.P)에 관한 것이다. 선택적으로, 본 발명은 기업 환경에서 기업 정책, 기밀 등급 및 준수 규칙을 추가로 적용하여 개인 기준과 기업 정책을 동시에 반영하는 이중 주권 기반 출력 제어를 제공한다. The present invention relates to a personalized artificial intelligence system that continues context without prompts when interacting with external artificial intelligence. Specifically, the present invention relates to a personalized artificial intelligence platform (A.I.P) in which a local clone artificial intelligence (A.I.C) deployed on a user terminal (PC, mobile, edge, or on-premise environment) does not output the inference results received from an external cloud-based large-scale artificial intelligence (large-scale language model or multimodal model) in their original form, but rather, by referring to the user's long-term context, judgment criteria, taboo conditions, and priority rules stored locally, selects, modifies, removes, or rearranges the inference results, and then reconstructs and provides them to the user. Optionally, in a corporate environment, the present invention provides dual sovereignty-based output control that simultaneously reflects personal criteria and corporate policies by additionally applying corporate policies, confidentiality levels, and compliance rules.","assignee":"이치헌","inventors":["이치헌"],"publication_date":"2026-03-03","filing_date":"2026-02-06","priority_date":"2026-02-06","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098","G","G06","G06F","G06F9/00","G06F9/06","G06F9/44"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260027991A/en"},{"publication_number":"MX2026001781A","title":"Configuring artificial intelligence (ai) bots with simulated personas for engaging in automated conversations","abstract":"Artificially intelligent (Al) bots with simulated personas can be used generate feedback about options. For example, a system can receive a selection by a chooser of an option from among a group of options. The system can configure, based on a chooser profile associated with the chooser, a first Al bot to simulate the chooser. The system can also configure, based on an end-user profile, a second Al bot to simulate an end user of the option. The system can then initiate a conversation about the selected option between the first Al bot and the second Al bot. Based on the conversation, the system can generate feedback about the option. The system can then provide the feedback about the option to the chooser.","assignee":"Kimberly Clark Co","inventors":["Tom M Ales","Stephen Becker","Jonathan D Boulos","Cleary E Mahaffey","Darren Parris","Mark Recio","Theodore T Tower"],"publication_date":"2026-03-02","filing_date":"2026-02-13","priority_date":"2023-09-06","cpc_codes":["H","H04","H04L","H04L51/00","H04L51/02","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","H","H04","H04L","H04L51/00","H04L51/21","H04L51/216"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2026001781A/en"},{"publication_number":"MX2026001698A","title":"Wireless communication method and communication device","abstract":"Provided are a wireless communication method and a communication device. The method comprises: a first device sends first information to a second device, the first information being used for managing a first model, and the first model being used for receiving a target signal. In embodiments of the present application, the first model can be managed by means of the first information, thereby avoiding the problem that a model at a sending end does not match a model at a receiving end.","assignee":"Guangdong Oppo Mobile Telecommunications Corp Ltd","inventors":["Han Xiao"],"publication_date":"2026-03-02","filing_date":"2026-02-11","priority_date":"2023-08-17","cpc_codes":["H","H04","H04L","H04L5/00","H04L5/003","H04L5/0058","H04L5/006","G","G06","G06N","G06N20/00","H","H04","H04L","H04L25/00","H04L25/02","H","H04","H04L","H04L5/00","H04L5/003","H04L5/0048","H04L5/005","H","H04","H04W","H04W24/00","H04W24/02"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2026001698A/en"},{"publication_number":"MX2026001305A","title":"Modeling plant ingredient proteins for use in developing food products","abstract":"Disclosed are systems and methods for determining a concentration of a plant ingredient in a developed food product before or while making the food product. A method may include: receiving, by a computing system, protein data for the plant ingredient, retrieving at least one model trained to generate output indicating a desired concentration of the plant ingredient to be used in developing the food product, the model including at least one of a pH, temperature, and protein-content model, providing the received protein data as input to the model, receiving the output from the model, and generating instructions for developing the food product including the desired concentration range of the plant ingredient.","assignee":"Frito Lay North America Inc","inventors":["Jenna Wang","Yi Zhu","Antonio Garay","Jingting Hui"],"publication_date":"2026-03-02","filing_date":"2026-01-30","priority_date":"2023-08-25","cpc_codes":["A","A23","A23J","A23J3/00","A23J3/14","A","A23","A23L","A23L11/00","G","G06","G06N","G06N3/00","G06N3/02","G","G16","G16B","G16B15/00","G16B15/20","G","G16","G16B","G16B40/00","G16B40/20"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2026001305A/en"},{"publication_number":"MX2026001131A","title":"Programmed food equilibration system and method for real time process yield in a thermal process","abstract":"A thermal processing apparatus for heating a first type of work products within an enclosure may include a first thermal processing zone operating under selected thermal operating parameters to heat the first type of work products to a predetermined maximum characteristic surface temperature not influenced by conditions external to a boundary of the first type of work products (TMCS); and, a programmed equilibration system, including: a second adiabatic equilibration zone defined within the enclosure of the thermal processing apparatus; and, a controller for substantially maintaining a temperature of the second adiabatic equilibration zone to a predetermined constant reference temperature (TR) that is substantially the same as the TMCS to enable the first type of work products to equilibrate adiabatically.","assignee":"Jbt Marel Corp","inventors":["Ramesh M Gunawardena","Owen E Morey"],"publication_date":"2026-03-02","filing_date":"2026-01-28","priority_date":"2023-08-02","cpc_codes":["A","A23","A23B","A23B2/00","A23B2/40","A23B2/42","A","A23","A23B","A23B2/00","A23B2/40","A","A23","A23L","A23L13/00","A","A23","A23L","A23L17/00","A","A23","A23L","A23L5/00","A23L5/10","A","A23","A23L","A23L5/00","A23L5/10","A23L5/17","G","G06","G06N","G06N20/00"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2026001131A/en"},{"publication_number":"MX2026001146A","title":"Quality factor estimation method, operation condition change method, model generation method, quality factor estimation device, and operation condition change device","abstract":"This quality factor estimation method comprises: a step for acquiring quality data indicating the quality of products for normal quality and m (m is equal to or greater than 2) types of quality defects; a step for acquiring operation data of a manufacturing process; a step for creating quality evaluation data for each of the m types of quality defects; a step for acquiring m types of models generated by using the quality evaluation data, and determining a variable importance indicating the magnitude of the contribution of the operation variable for each of the m types of models; and a step for identifying an operation variable estimated to be the main factor with respect to at least one of the m types of quality defects by comparing the variable importances of the m types of models.","assignee":"Jfe Steel Corp","inventors":["Hiroshi Nakagawa","Hiroyuki Takagi","Shuji Kuyama","Takehide Hirata"],"publication_date":"2026-03-02","filing_date":"2026-01-28","priority_date":"2023-07-31","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06Q","G06Q50/00","G06Q50/04","Y","Y02","Y02P","Y02P90/00","Y02P90/30"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2026001146A/en"},{"publication_number":"MX2026000636A","title":"Multi-layer split points output information","abstract":"A WTRU may perform inference processing on video data to generate intermediate data. The WTRU may determine from the intermediate data a plurality of tuples and may generate metadata from the plurality of tuples. The metadata may comprise an encoding type that may indicate an encoding algorithm. The metadata may further comprise a length indicating the length of the metadata. The metadata may also comprise an indication of the number of tuples that are comprised in the metadata. The metadata may further comprise the plurality of tuples. Each tuple may comprise a respective layer identifier and tensor shape information. The device may generate a bitstream from the intermediate data and may transmit the bitstream and the metadata to another device which may perform split inference processing on the generated bitstream using the metadata.","assignee":"Interdigital Vc Holdings Inc","inventors":["Stephane Onno","Fabien Racape","Cyril Quinquis","Thierry Filoche"],"publication_date":"2026-03-02","filing_date":"2026-01-15","priority_date":"2023-07-19","cpc_codes":["H","H04","H04N","H04N19/00","H04N19/70","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2026000636A/en"},{"publication_number":"JP2026034836A","title":"Information processing device, information processing method, and program","abstract":"【課題】ロボット装置の制御に関して、説明性の高い出力を得る技術を提供する。【解決手段】本発明の一側面に係る情報処理装置は、ロボット装置の動作する環境の観測データ、及びロボット装置に与えるタスクの目標に関する指示情報を取得し、推論モジュールを用いて、取得された観測データ及び指示情報からタスクの問題記述を生成し、かつ生成された問題記述を出力する。問題記述は、環境に存在する物体の初期状態及び目標状態の記述を含むように構成される。【選択図】図１ [Problem] To provide a technology that can obtain highly explainable output regarding the control of a robot device. [Solution] An information processing device according to one aspect of the present invention acquires observation data of the environment in which the robot device operates and instruction information regarding the objective of the task to be given to the robot device, generates a problem description of the task from the acquired observation data and instruction information using an inference module, and outputs the generated problem description. The problem description is configured to include a description of the initial state and target state of objects present in the environment. [Selected Figure] Figure 1","assignee":"Omron Corp","inventors":["政志 ▲濱▼屋","敦史 橋本","翔平 田中","圭佑 白井","カミロ ベルトラン エルナンデス クリスティアン"],"publication_date":"2026-03-02","filing_date":"2025-12-16","priority_date":"2023-10-25","cpc_codes":["B","B25","B25J","B25J13/00","G","G06","G06N","G06N20/00","G","G06","G06N","G06N5/00","G06N5/04","G06N5/045"],"country":"JP","kind":"application","source_url":"https://patents.google.com/patent/JP2026034836A/en"},{"publication_number":"CA3282602A1","title":"Status and monitoring platform","abstract":"A status and monitoring platform for resource bandwidth is provided. A system can retrieve, responsive to a request for bandwidth of a resource, a data set that can include at least one constraint related to the resource and historic utilization of the resource. The system can construct, based on the data set, a data structure to replace the request. Based on the data structure, the system can generate a prompt indicating the constraint and the historic utilization. The system can identify, based on the prompt, a model trained with generative artificial intelligence to determine resource bandwidth. The system can input the prompt into the model to generate an output that indicates the bandwidth of the resource and validate the output based on a comparison with a threshold. The system can transmit for display, via an interface, responsive to the validation, an indication of resource bandwidth output by the model.","assignee":"ADP Inc","inventors":["Savitri Katam","Bhavani Meegada","Monika Nagalla","Haneesh Bathini","Pavan Kumar Telluri","Parag Khare","Harsh Singh"],"publication_date":"2026-03-01","filing_date":"2025-08-11","priority_date":"2024-08-12","cpc_codes":["H","H04","H04L","H04L41/00","H04L41/14","G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/105","G06Q10/1057","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06313","G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/105","H","H04","H04L","H04L41/00","H04L41/16","H","H04","H04L","H04L41/00","H04L41/22","H","H04","H04L","H04L43/00","H04L43/08","H04L43/0876"],"country":"CA","kind":"application","source_url":"https://patents.google.com/patent/CA3282602A1/en"},{"publication_number":"CA3250435A1","title":"Smart contract configuration using unstructured data","abstract":"Systems and methods for configuring a smart contract are disclosed. A system may receive unstructured data and identifying, based on the unstructured data, an intent to perform a transfer of data. The system may then identify, based on the unstructured data, parameters associated with the transfer of data. The parameters may include an amount of data to be transferred, a transferee, and at least one condition associated with the transfer of data. The system may then retrieve identifier data from a storage module, the identifier data including a first unique identifier. The system may then send, to a Large Language Model (LLM) via a first prompt engine module and an LLM Application Programming Interface (API), a prompt based on the unstructured data and the identifier data. The system may then receive LLM output from the LLM, and based on the LLM output, configure a smart contract.","assignee":"Toronto Dominion Bank","inventors":["Hitesh Bajaj","Milos Dunjic","David Samuel TAX","Jonathan Joseph PRENDERGAST","Mohit Sharma","Thomas Osman KELLY","Pranay Chander GUPTA","Kushank RASTOGI","Abhijit SINGHA HAZARI"],"publication_date":"2026-03-01","filing_date":"2024-07-31","priority_date":"2024-07-11","cpc_codes":["G","G06","G06F","G06F21/00","G06F21/60","G06F21/64","G","G06","G06F","G06F40/00","G06F40/20","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2455","G","G06","G06F","G06F16/00","G06F16/20","G06F16/28","G06F16/284","G06F16/285","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/90","G","G06","G06F","G06F21/00","G06F21/50","G06F21/55","G06F21/56","G","G06","G06F","G06F21/00","G06F21/50","G06F21/57","G06F21/577","G","G06","G06F","G06F2221/00","G06F2221/03","G06F2221/034","G","G06","G06F","G06F40/00","G06F40/40","G","G06","G06F","G06F40/00","G06F40/40","G06F40/55","G06F40/56","G","G06","G06N","G06N20/00"],"country":"CA","kind":"application","source_url":"https://patents.google.com/patent/CA3250435A1/en"},{"publication_number":"CN121579583A","title":"A method and system for identifying and standardizing construction project cost data","abstract":"本发明公开了一种建筑工程成本数据识别与标准化方法及系统，包括采集并预处理多源异构成本数据得到初始数据集；构建五级成本数据标签体系与集成领域规则的成本知识图谱；基于标签体系标注历史宽表生成训练样本，训练深度学习模型得到标签匹配模型；将初始数据集映射为标准宽表后输入模型完成字段自动化标注；基于知识图谱规则对标注宽表进行字段值校验与数据归一化处理；输出标准化成本宽表；通过审核结果反馈至模型训练过程形成闭环优化。本发明通过五级标签体系与知识图谱的结合，实现了成本数据的自动化、标准化处理，显著提升了数据质量与跨项目可比性，为建筑工程成本管理提供可靠数据支撑。 This invention discloses a method and system for identifying and standardizing construction engineering cost data. The method includes: collecting and preprocessing multi-source heterogeneous cost data to obtain an initial dataset; constructing a five-level cost data labeling system and an integrated domain rule-based cost knowledge graph; generating training samples based on historical wide tables labeled with the labeling system, and training a deep learning model to obtain a label matching model; mapping the initial dataset to a standard wide table and inputting it into the model to complete automated field labeling; performing field value verification and data normalization on the labeled wide table based on knowledge graph rules; outputting a standardized cost wide table; and feeding back the verification results to the model training process to form a closed-loop optimization. This invention, through the combination of a five-level labeling system and a knowledge graph, achieves automated and standardized processing of cost data, significantly improving data quality and cross-project comparability, and providing reliable data support for construction engineering cost management.","assignee":"Hangzhou Ruicheng Information Technology Co ltd","inventors":["董维浩","嵇翔","高志瑞"],"publication_date":"2026-02-27","filing_date":"2026-01-29","priority_date":"2026-01-29","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/25","G06F16/254","G","G06","G06F","G06F16/00","G06F16/20","G06F16/21","G06F16/215","G","G06","G06F","G06F16/00","G06F16/20","G06F16/22","G","G06","G06F","G06F16/00","G06F16/20","G06F16/25","G06F16/258","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G","G06","G06Q","G06Q50/00","G06Q50/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121579583A/en"},{"publication_number":"CN121583147A","title":"Multimodal spatiotemporal fusion intelligent vehicle hazard prediction method and system","abstract":"本发明涉及智能交通和辅助出行技术领域，公开一种多模态时空融合智能车辆危险预判方法及系统，包括：获取智能车辆周围的多模态传感数据，构建基于深度学习的危险预判模型，模型包括时空融合模块和风险演化模块；时空融合模块时空对齐和融合多模态传感数据得到统一的时空体素特征，通过风险注意力门控模块使时空对齐和融合处理过程聚焦于高不确定度区域，风险演化模块根据时空体素特征识别智能车辆周围的物体、预测物体的运动轨迹和潜在碰撞风险；使用世界模型蒸馏危险预判模型，使用蒸馏后的危险预判模型对智能车辆进行实时危险判断。本发明可以有效融合多种传感数据进行车辆行驶过程中的危险判断，提高判断准确性。 This invention relates to the field of intelligent transportation and assisted mobility technology, and discloses a multimodal spatiotemporal fusion method and system for hazard prediction of intelligent vehicles. The method includes: acquiring multimodal sensor data around the intelligent vehicle; constructing a deep learning-based hazard prediction model, which includes a spatiotemporal fusion module and a risk evolution module; the spatiotemporal fusion module aligns and fuses the multimodal sensor data to obtain unified spatiotemporal voxel features; a risk attention gating module focuses the spatiotemporal alignment and fusion process on high-uncertainty regions; the risk evolution module identifies objects around the intelligent vehicle based on the spatiotemporal voxel features, predicts the motion trajectories of the objects, and identifies potential collision risks; the hazard prediction model is distilled using a world model; and the distilled hazard prediction model is used to perform real-time hazard assessment of the intelligent vehicle. This invention can effectively fuse multiple sensor data for hazard assessment during vehicle operation, improving the accuracy of the assessment.","assignee":"Suzhou Guanrui Automobile Technology Co ltd","inventors":["丁延超","陈轶宁","胡翔睦","贺冬","陈赛","田欢"],"publication_date":"2026-02-27","filing_date":"2026-01-29","priority_date":"2026-01-29","cpc_codes":["G","G08","G08G","G08G1/00","G08G1/16","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/254","G06F18/256","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G08","G08B","G08B31/00","G","G08","G08G","G08G1/00","G08G1/01","G08G1/0104","G08G1/0108","G08G1/012","G","G08","G08G","G08G1/00","G08G1/01","G08G1/0104","G08G1/0125"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121583147A/en"},{"publication_number":"CN121581340A","title":"A precise method for calculating the carbon footprint of lithium-ion batteries based on regional energy structure","abstract":"The invention discloses a lithium ion battery carbon footprint accurate accounting method based on regional energy structures, relates to the technical field of data accounting, and aims to determine energy structures and corresponding duty ratios in all regions, acquire unit carbon emission factors corresponding to the energy structures based on a database, calculate the unit carbon emission factors to obtain comprehensive electric power carbon emission factors in all regions, construct a comprehensive electric power carbon emission factor model, and construct an accounting stage model of a full life cycle. According to the invention, the region corresponding to the comprehensive electric power carbon emission factor can be marked and stored through the comprehensive electric power carbon emission factor model, the carbon emission amount of the lithium ion battery can be rapidly and accurately calculated based on the comprehensive electric power carbon emission factor model and the energy consumption data of the lithium ion battery in each stage and region, and the calculation of the carbon footprint under the full life cycle of the lithium ion battery by the sub-regions is more accurate than the calculation according to the fixed factor.","assignee":"Chongqing University","inventors":["李军","刘兰"],"publication_date":"2026-02-27","filing_date":"2026-01-29","priority_date":"2026-01-29","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/067"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121581340A/en"},{"publication_number":"CN121585874A","title":"A Zero-Shot Video Editing Method and System Based on Adaptive Segmentation and Cross-Segment Interaction","abstract":"The invention belongs to the technical field of video editing, and particularly relates to a zero sample video editing method and system based on self-adaptive segmentation and cross-segment interaction, wherein the method comprises the steps of encoding a source video into a latent space and extracting characteristics to construct an initial noise latent variable; the method comprises the steps of adaptively dividing semantic segments based on feature similarity, selecting representative anchor frames, screening redundant Token in space-time adjacent domains based on the anchor frames, carrying out soft weighting combination, synchronously constructing a mapping index table, alternately combining Token sets according to double modes in the later period of diffusion, carrying out cross-segment joint self-attention calculation based on the index table to enhance global continuity, and finally carrying out iterative denoising in a collaborative attention mechanism to generate an edited video maintained by a time sequence structure. The invention realizes the time sequence structure preservation, cross-segment semantic coherence and calculation efficiency improvement in long video editing, and can finish high-quality video editing without model fine tuning.","assignee":"Anhui University","inventors":["刘德银","丁义胜","金哲","都浩","魏李武","张�浩"],"publication_date":"2026-02-27","filing_date":"2026-01-29","priority_date":"2026-01-29","cpc_codes":["H","H04","H04N","H04N21/00","H04N21/40","H04N21/47","H04N21/472","H04N21/47205","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","H","H04","H04N","H04N21/00","H04N21/20","H04N21/23","H04N21/234","H04N21/23412","H","H04","H04N","H04N21/00","H04N21/20","H04N21/23","H04N21/234","H04N21/23418","H","H04","H04N","H04N21/00","H04N21/40","H04N21/43","H04N21/44","H04N21/44008","H","H04","H04N","H04N21/00","H04N21/40","H04N21/43","H04N21/44","H04N21/44012","H","H04","H04N","H04N21/00","H04N21/80","H04N21/83","H04N21/845"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121585874A/en"},{"publication_number":"CN121585483A","title":"Methods for generating code vulnerability mitigation strategies and related equipment","abstract":"The disclosure provides a method and related equipment for generating a coping strategy of a code vulnerability, wherein the method comprises the steps of responding to a log record instruction, generating a target log query statement according to the type of the log record instruction, wherein the target log query statement is used for querying a log related to log4j vulnerability attack behavior characteristics, executing the target log query statement to obtain a target log query result, judging whether the log record instruction contains the suspected log4j vulnerability attack behavior according to the target log query result, if so, carrying out link identification on the log4j vulnerability attack to obtain a target attack link, and generating the target coping strategy for the log4j vulnerability attack according to the target attack link. The present disclosure may improve network security defenses.","assignee":"Shanghai Douxiang Information Technology Co ltd","inventors":["徐钟豪","陈伟","李非凡"],"publication_date":"2026-02-27","filing_date":"2026-01-29","priority_date":"2026-01-29","cpc_codes":["H","H04","H04L","H04L63/00","H04L63/14","H04L63/1433","G","G06","G06F","G06F21/00","G06F21/50","G06F21/57","G06F21/577","H","H04","H04L","H04L63/00","H04L63/14","H04L63/1408","H04L63/1425","H","H04","H04L","H04L9/00","H04L9/40","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121585483A/en"},{"publication_number":"CN121581827A","title":"Knowledge Graph-Based Human Resource Record Storage Management System","abstract":"本发明涉及人力资源管理技术领域，具体为基于知识图谱的人力资源档案储存管理系统，系统包括：岗位识别模块、路径建模模块、节点联动模块、断裂识别模块、访问控制模块。本发明中，通过对岗位描述与技能组合的归一映射，提升非结构化档案内容的标准化处理能力，结合任务记录中多阶段路径节点的跳转频次与跨度特征，构建贴合岗位职责流程的阶段标签体系，利用任职周期划分结果识别节点间责任字段的交叉关联，动态调整路径结构映射顺序，精准识别结构路径中职责传导关系的变更情况，并在连续空岗路径识别中通过字段组差异判断责任链中断范围，基于访问路径中敏感字段分布特征及角色字段覆盖比例，实现路径级别的访问许可判断与权限校核控制。 This invention relates to the field of human resource management technology, specifically a knowledge graph-based human resource file storage and management system. The system includes: a job identification module, a path modeling module, a node linkage module, a break identification module, and an access control module. In this invention, by normalizing the mapping between job descriptions and skill combinations, the standardization processing capability of unstructured file content is improved. Combining the jump frequency and span characteristics of multi-stage path nodes in task records, a stage label system that fits the job responsibility process is constructed. The cross-correlation of responsibility fields between nodes is identified using the results of the tenure cycle division, dynamically adjusting the path structure mapping order, accurately identifying changes in the responsibility transmission relationship in the structured path, and determining the scope of responsibility chain interruption in continuous vacancy path identification through field group differences. Based on the distribution characteristics of sensitive fields in the access path and the coverage ratio of role fields, path-level access permission judgment and authorization verification control are achieved.","assignee":"Guizhou Bluesky Innovative Science & Technology Co ltd","inventors":["吴利平","杨福贵","杨丽盼","张亚琳","张向志","艾刚"],"publication_date":"2026-02-27","filing_date":"2026-01-29","priority_date":"2026-01-29","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/105","G","G06","G06F","G06F40/00","G06F40/20","G06F40/237","G06F40/247","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121581827A/en"},{"publication_number":"CN121579961A","title":"A Virtual Sample Generation Method and System Based on Non-Stationary Neural Network Gaussian Processes","abstract":"The invention provides a virtual sample generation method and a system based on a non-stationary neural network Gaussian process, wherein the method comprises the steps of obtaining original data, judging non-stationary property, judging whether statistical properties change along with input positions, if not stationary, constructing an hidden variable model fusing the neural network and the Gaussian process, training to obtain a hidden variable space representing non-stationary distribution, and sampling based on hidden variable space probability distribution Generating by model through non-stable high-dimensional mapping = ( ,Z) (Z, Z) ‑1 X, where Z is a low-dimensional hidden variable, and X is an input variable; the method can generate high-quality and high-diversity virtual samples under the conditions of small samples and non-stable scenes.","assignee":"Central South University","inventors":["胡敏","陈凯"],"publication_date":"2026-02-27","filing_date":"2026-01-29","priority_date":"2026-01-29","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/211","G06F18/2111","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N3/00","G06N3/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121579961A/en"},{"publication_number":"AU2026200822A1","title":"Gaming activity monitoring systems and methods","abstract":"A system for monitoring gaming activity in a gaming area with a gaming table, comprising a camera configured to capture images of the gaming area, a processor to communicate with the camera and a memory, the memory storing instructions executable by the processor to configure the processor to determine a presence of a first gaming object on the gaming table in a first image from a series of images of the gaming area captured by the camera, the series of images comprising one or more images, responsive to determining the presence of the first gaming object in the first image, process the first image to estimate postures of one or more players in the first image, and based on the estimated postures, determine a first target player associated with the first gaming object among the one or more players. WO 2022/000023 PCT/AU2021/0505921/13 Gaming Monitoring Computing Device 120 100 Memory 124 Object Detection Module 123 Wager Estimation Game Object Module 126 Detection NN 151 Camera 110 Height based Person wager estimation Detection NN sub-module 154 159 Pose Estimation Edge Module 125 Recognition Camera 112 NN 155 Key point Estimation NN 152 Face Recognition Module 127 3D Mapping Gaming Table 140 NN 153 Face Embedding Generation NN Game Object 156 Association Module 128 Processor 122 Gaming Monitoring Server 180 Network Interface 129 Processor 182 Memory 184 Network 130 Facial Feature and Identity Database 189 Figure 1 WO 2022/000023 PCT/AU2021/050592 100 Memory 124 Module 123 Module 126 Detection NN 151 Camera 110 Person Detection NN sub-module 154 159 Pose Estimation Module 125 Camera 112 NN 155 Estimation NN 152 Module 127 NN 153 Generation NN 156 Module 128 Processor 122 180 Network Interface 129 Processor 182 Network 130 Database 189 20 26 20 08 22 05 F eb 2 02 6 w o 2 0 2 2 / 0 0 0 0 2 3 P C T / A U 2 0 2 1 / 0 5 0 5 9 2 2 0 2 6 2 0 0 8 2 2 0 5 F e b 2 0 2 6 M o d u l e 1 2 6 D e t e c t i o n N N 1 5 1 C a m e r a 1 1 0 P e r s o n D e t e c t i o n N N s u b - m o d u l e 1 5 4 1 5 9 P o s e E s t i m a t i o n M o d u l e 1 2 5 C a m e r a 1 1 2 N N 1 5 5 E s t i m a t i o n N N 1 5 2 M o d u l e 1 2 7 3 D M a p p i n g N N 1 5 3 G e n e r a t i o n N N 1 5 6 M o d u l e 1 2 8 P r o c e s s o r 1 2 2 1 8 0 N e t w o r k I n t e r f a c e 1 2 9 P r o c e s s o r 1 8 2 M e m o r y 1 8 4 N e t w o r k 1 3 0 D a t a b a s e 1 8 9","assignee":"Angel Group Co Ltd","inventors":["Subhash Challa","Louis Quinn","Duc Dinh Minh Vo","Nhat Vo"],"publication_date":"2026-02-26","filing_date":"2026-02-05","priority_date":"2020-06-30","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G","G06","G06T","G06T7/00","G06T7/0002","G","G06","G06T","G06T7/00","G06T7/10","G","G06","G06T","G06T7/00","G06T7/70","G06T7/73","G","G06","G06T","G06T7/00","G06T7/70","G06T7/73","G06T7/75","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/40","G06V20/41","G06V20/42","G","G06","G06V","G06V20/00","G06V20/40","G06V20/44","G","G06","G06V","G06V20/00","G06V20/50","G06V20/52","G","G06","G06V","G06V20/00","G06V20/60","G","G06","G06V","G06V20/00","G06V20/70","G","G06","G06V","G06V40/00","G06V40/10","G","G06","G06V","G06V40/00","G06V40/10","G06V40/103","G","G06","G06V","G06V40/00","G06V40/10","G06V40/107","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G06V40/161","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G06V40/168","G06V40/171","G","G06","G06V","G06V40/00","G06V40/20","G06V40/28","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3202","G07F17/3204","G07F17/3206","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3202","G07F17/3223","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3225","G07F17/3232","G07F17/3234","G","G07","G07F","G07F17/00","G07F17/32"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026200822A1/en"},{"publication_number":"AU2026200760A1","title":"Stock rewards in consumer transactions","abstract":"A method that includes receiving a string from a remote station, the string associated with an interaction event between a user and the remote station, is provided. The method includes verifying that the string includes a content validation for a user account in a network service, mapping at least a portion of the string to a ticker symbol associated with an entry in a database and transmitting, to the user, a message indicating that a fractional value associated with the ticker symbol has been added to the user account in the network service. A system configured to execute the above method is also provided.","assignee":"Stash Financial Inc","inventors":["Michael BARANY","Adam Finley","Clifford HAZELTON","Brandon KRIEG","Edward Robinson","Kyle SCHUSTAK","Evan Weiss"],"publication_date":"2026-02-26","filing_date":"2026-02-03","priority_date":"2019-03-05","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/10","G","G06","G06Q","G06Q10/00","G06Q10/10","G","G06","G06Q","G06Q20/00","G06Q20/08","G06Q20/12","G06Q20/127","G","G06","G06Q","G06Q20/00","G06Q20/08","G06Q20/20","G06Q20/202","G","G06","G06Q","G06Q20/00","G06Q20/30","G06Q20/32","G06Q20/322","G","G06","G06Q","G06Q20/00","G06Q20/38","G06Q20/387","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0207","G06Q30/0213","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0207","G06Q30/0215","G06Q30/0216","G","G06","G06Q","G06Q40/00","G06Q40/04","G","G07","G07F","G07F17/00","G07F17/0014","G07F17/0035","H","H04","H04L","H04L63/00","H04L63/12","H04L63/123"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026200760A1/en"},{"publication_number":"AU2026200755A1","title":"Database management and graphical user interfaces for measurements collected by analyzing blood","abstract":"- 101 - Methods and devices include database management and graphical user interfaces for measurements collected by analyzing blood.","assignee":"WellDoc Inc","inventors":["Anand Iyer","Hari KESANI","Kevin MCRAITH","Prasad Matti Rao","Mansur Shomali","Gabriel SUSAI"],"publication_date":"2026-02-26","filing_date":"2026-02-02","priority_date":"2016-05-13","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/20","A","A61","A61B","A61B5/00","A61B5/145","A61B5/14503","A","A61","A61B","A61B5/00","A61B5/145","A61B5/14532","G","G06","G06N","G06N20/00","G","G16","G16H","G16H10/00","G16H10/20","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H20/00","G","G16","G16H","G16H20/00","G16H20/10","G","G16","G16H","G16H20/00","G16H20/60","G","G16","G16H","G16H40/00","G16H40/60","G16H40/67"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026200755A1/en"},{"publication_number":"KR20260025343A","title":"Electronic device for distributing datasets for lightweight artificial intelligence model based on on-device","abstract":"온 디바이스 기반의 경량화 인공지능 모델을 위한 데이터셋을 분배하는 전자 장치가 개시된다. 본 발명의 전자 장치는, 데이터셋에 포함된 복수의 하위 항목들에 대하여 복수의 속성값들을 정의하고 저장하는 속성 정의부, 속성 정의부에 의해 정의된 하위 항목들을, 각 하위 항목에 상응하는 상위분류 단위로 그룹화하고, 각 상위분류 단위에 포함된 하위 항목들을 하나의 모델에만 배정되도록 고정하는, 상위분류 그룹 생성부, 미리 설정된 모델의 수를 기준으로, 각 모델 간의 부하량 차이를 산출하여, 부하량 차이를 최소화하도록 각 상위분류 단위를 모델에 할당하는, 부하 균형 기반 배정부, 상위분류 단위의 모델 할당에 따른 부하량 차이, 유사도 충돌량 및 다양성 수치의 변화량을 기반으로 분배 평가 기준값을 산출하는, 분배 기준값 평가부, 분배 평가 기준값을 기반으로, 각 상위분류 단위가 배정될 수 있는 복수의 모델 중에서, 분배 평가 기준값이 가장 작은 값이 산출되는 모델로 상위분류 단위를 반복적으로 재배정하여, 미리 설정된 임계값 이하의 평가 기준값이 산출되는 모델을 최종 할당 대상으로 선택하는, 모델 재배정 제어부를 포함한다. An electronic device for distributing a dataset for an on-device based lightweight artificial intelligence model is disclosed. The electronic device of the present invention includes an attribute definition unit that defines and stores a plurality of attribute values for a plurality of sub-items included in a dataset, an upper classification group generation unit that groups the sub-items defined by the attribute definition unit into upper classification units corresponding to each sub-item and fixes the sub-items included in each upper classification unit to be assigned to only one model, a load balance-based assignment unit that calculates a load difference between each model based on a preset number of models and assigns each upper classification unit to a model so as to minimize the load difference, a distribution criterion evaluation unit that calculates a distribution evaluation criterion based on the load difference, similarity conflict amount, and diversity value change according to model assignment of the upper classification unit, and a model reassignment control unit that repeatedly reassigns the upper classification unit to a model that calculates a value having a smallest distribution evaluation criterion value among a plurality of models to which each upper classification unit can be assigned based on the distribution evaluation criterion value and selects a model that calculates an evaluation criterion value below a preset threshold value as a final assignment target.","assignee":"주식회사 알테르모","inventors":["석준현","고정규","이종석"],"publication_date":"2026-02-24","filing_date":"2026-02-05","priority_date":"2025-06-30","cpc_codes":["G","G06","G06N","G06N20/00","G06N20/20"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260025343A/en"},{"publication_number":"KR20260025335A","title":"Multi-Criteria Conflict Auto-Adjustment Engine System","abstract":"본 발명은 산업 설비, 작업 환경 및 각종 시스템에서 발생하는 간접 계측 신호 및 외부 기준 데이터를 이용하여 이상 징후를 조기에 검출하고, 복수의 판단 기준을 자동 생성한 후 기준 간 충돌을 정량적으로 분석 및 자동 조정하여 통합 예방 판단을 수행하는 시스템 및 방법에 관한 것이다. 본 발명에 따르면, 간접 계측 신호 또는 외부 기준 데이터로부터 수집된 질환관리기준, 산업계측기준, 규제기준 및 운영정책기준을 기준수집모듈이 수집하고, 파라미터변환모듈이 이를 임계값 벡터, 허용범위 모델, 위험 가중치 및 성능 지표를 포함하는 수치기준 세트로 변환한다. 충돌분석엔진은 상기 수치기준 세트에 대해 중첩영역, 상충영역 및 위반빈도를 분석하여 충돌강도와 충돌리스트를 산출하고, 자동조정엔진은 상기 충돌강도에 기초하여 기준 파라미터를 자동 조정하여 조정 파라미터 세트를 생성한다. 통합기준생성모듈은 조정된 기준 파라미터로부터 통합기준을 생성하며, 예방판단출력모듈은 상기 통합기준에 따라 위험도 점수, 경보 및 제어 또는 권고안을 포함하는 예방판단 결과를 출력한다. 이에 따라본 발명은 다중 기준 간 충돌 문제를 자동으로 해소하면서 환경 변화에 적응하는 예방 판단을 구현하여, 이상 징후의 조기 대응, 안전성 향상, 운영 효율 증대 및 관리 비용 절감을 동시에 달성할 수 있는 효과를 제공한다. The present invention relates to a system and method for early detection of abnormal signs using indirect measurement signals and external reference data generated in industrial facilities, work environments, and various systems, automatically generating multiple judgment criteria, and then quantitatively analyzing and automatically adjusting conflicts between criteria to perform integrated preventive judgment. According to the present invention, a criteria collection module collects disease management criteria, industrial measurement criteria, regulatory criteria, and operational policy criteria collected from indirect measurement signals or external reference data, and a parameter conversion module converts them into a numerical criteria set including a threshold vector, an allowable range model, a risk weight, and a performance indicator. A conflict analysis engine analyzes the overlapping area, conflict area, and violation frequency of the numerical criteria set to derive a conflict intensity and a conflict list, and an automatic adjustment engine automatically adjusts the criteria parameters based on the conflict intensity to generate an adjustment parameter set. An integrated criteria generation module generates an integrated criteria from the adjusted criteria parameters, and a prevention judgment output module outputs a prevention judgment result including a risk score, an alert, and a control or recommendation according to the integrated criteria. Accordingly, the present invention provides the effect of simultaneously achieving early response to abnormal signs, improved safety, increased operational efficiency, and reduced management costs by implementing preventive judgment that adapts to environmental changes while automatically resolving conflict issues between multiple criteria.","assignee":"구교선; 구현우","inventors":["구교선","구현우"],"publication_date":"2026-02-24","filing_date":"2026-02-05","priority_date":"2026-02-05","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06316","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0633","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G","G08","G08B","G08B21/00","G08B21/18"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260025335A/en"},{"publication_number":"CN121563022A","title":"Multi-mode data fusion-based orchard single fruit picking optimization method","abstract":"The invention relates to an orchard single fruit picking optimization method based on multi-mode data fusion, which is characterized in that image information, local environment data and tree physiological indexes of fruits are synchronously acquired through a multi-mode sensor, the image information, the local environment data and the tree physiological indexes are preprocessed, deep features of the fruits are acquired by adopting independent feature extraction networks aiming at heterogeneous characteristics of different mode data, a heterogeneous mode depth fusion network is constructed, complex correlations affecting fruit maturity are effectively captured through cross-domain attention mechanism dynamic integration features, fusion features are generated, real-time maturity of the single fruits can be predicted by the model based on the features, and the picking priority score of each fruit is calculated by utilizing a multi-target optimization algorithm in combination with external conditions, so that an executable picking sequence is generated, and a robot or a person is guided to conduct orderly picking. The method and the device remarkably improve the accuracy of maturity prediction and the comprehensive benefit of picking operation, and provide effective support for realizing the fine management and intelligent operation of the orchard.","assignee":"CETC Big Data Research Institute Co Ltd","inventors":["黄江","谢真强","张坤勇","苑建坤","余楷","郑荣华","胡婷","陈然","龙曦","曹扬","谢红韬","支婷","蔡惠民","管桂林","汪洋舟","陶政坪","丁志","王清青","王序","刘铭","李丙林","龙宇","李中坤","卢元东","欧政远","李宇航","雷参愉","李夏强","丁洪鑫","杨书","陈辉","燕熙迪","刘晓军"],"publication_date":"2026-02-24","filing_date":"2026-01-26","priority_date":"2026-01-26","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q50/00","G06Q50/02","G","G06","G06V","G06V10/00","G06V10/20","G06V10/26","G","G06","G06V","G06V20/00","G06V20/10","G06V20/194"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121563022A/en"},{"publication_number":"CN121563274A","title":"Mine safety decision deduction method based on digital twinning","abstract":"The invention discloses a mine safety decision deduction method based on digital twinning, which relates to the technical field of mining engineering and comprises the steps of constructing a three-dimensional geometric mine structure model based on a multi-source fusion data set, carrying out attribute injection and discretization on the three-dimensional geometric mine structure model to generate a real-time mine digital twinning body, defining mine element nodes and relationship sides based on the multi-source fusion data set, constructing a safety-working procedure-resource knowledge graph, acquiring mine state fields from the safety-working procedure-resource knowledge graph and the real-time mine digital twinning body, constructing a mine safety decision state space and an action space, identifying decision disturbance events in multi-source monitoring data according to a trigger threshold, and combining the mine safety decision state space and the action space to construct a mine safety decision rewarding function. The invention improves the semantic reasoning capacity of the decision process and achieves the effects of improving the decision accuracy, the instantaneity and the intelligent predictability.","assignee":"Guizhou Mine Safety Research Institute Co ltd; Guizhou University","inventors":["李青松","张伟","蒋星星","张义平","魏晏军","郑禄林","张书金","张朋","申振华","段正鹏","张勇"],"publication_date":"2026-02-24","filing_date":"2026-01-26","priority_date":"2026-01-26","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/254","G06F18/256","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/067","G","G06","G06Q","G06Q50/00","G06Q50/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121563274A/en"},{"publication_number":"CN121563447A","title":"Consumer equity protection intelligent examination method and system based on large language model","abstract":"The invention provides a consumer interest protection intelligent examination method and system based on a large language model, which relate to the technical field of artificial intelligence and natural language processing, and the method comprises the steps of preprocessing a text to be examined to obtain structured text characteristic data; the method comprises the steps of carrying out semantic understanding on structured text feature data by utilizing a pre-trained large language model to obtain text semantic vectors containing field semantic information, carrying out joint retrieval and matching on the text semantic vectors by combining a knowledge graph and a rule base, identifying and outputting a risk element set, and carrying out compliance judgment on the basis of the risk element set to obtain a preliminary audit conclusion containing risk level, evidence sufficiency, rule matching degree and reference confidence degree. According to the method, through intelligent processing of the whole flow, the original text is automatically converted into the high-reliability audit conclusion with the interpretability.","assignee":"Shanghai Rongshu Information Technology Co ltd","inventors":["刘璐","赖建章","李�荣","林永","刘迎春","孙祥"],"publication_date":"2026-02-24","filing_date":"2026-01-26","priority_date":"2026-01-26","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/103","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3344","G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","Y","Y02","Y02P","Y02P90/00","Y02P90/30"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121563447A/en"},{"publication_number":"CN121558642A","title":"Intelligent farmland soil humidity monitoring and early warning system based on multispectral remote sensing","abstract":"The invention relates to the technical field of agricultural information, and particularly discloses an intelligent farmland soil humidity monitoring and early warning system based on multispectral remote sensing. And generating a continuous characteristic sequence through multi-source data fusion, performing Jie Oupu adaptability and regional characteristics to adapt to a new scene, performing high-precision inversion in cooperation with multi-scale information, and performing anomaly early warning and cause tracing based on a dynamic threshold. The invention realizes all-weather continuous soil humidity monitoring, improves the generalization capability and monitoring precision of the model in different farmland areas, and provides intelligent decision support for agricultural irrigation management.","assignee":"Shaanxi Agriculture And Forestry Vocational And Technical University","inventors":["董拴涛"],"publication_date":"2026-02-24","filing_date":"2026-01-26","priority_date":"2026-01-26","cpc_codes":["G","G01","G01N","G01N21/00","G01N21/17","G01N21/25","G","G01","G01N","G01N33/00","G01N33/24","G01N33/245","G","G01","G01N","G01N33/00","G01N33/24","G01N33/246","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/23","G06F18/232","G06F18/2321","G06F18/23213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2433","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N5/00","G06N5/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121558642A/en"},{"publication_number":"CN121559900A","title":"Intelligent supercritical CO2 injection regulation and control method based on AI model","abstract":"The invention discloses an intelligent regulation and control method for supercritical CO2 injection based on an AI model, which relates to the technical field of fluid injection regulation and control, and comprises the steps of collecting real-time multi-source monitoring data, constructing a high-precision digital twin model, inputting the real-time multi-source monitoring data into the high-precision digital twin model for dynamic simulation, and outputting an underground pressure field and a CO2 concentration field; the control instruction set is issued to an electric regulating valve and a high-pressure pump of the injection well and a pumping pump of the pumping well, the injection pressure, the injection flow rate and the pumping pump power of CO2 are regulated and controlled to generate a regulation event record, and based on the regulation event record, updated real-time multi-source monitoring data are collected and fed back to a high-precision digital twin model to form closed-loop regulation. According to the invention, through the variation assimilation framework of the integrated reinforcement learning agent, the on-line correction of model parameters is realized, and the real-time response capability and regulation accuracy to the underground state dynamic change are improved.","assignee":"Guizhou Mine Safety Research Institute Co ltd; Guizhou University","inventors":["李青松","魏晏军","郑禄林","张伟","张义平","任富强","蒋星星","张书金","王春华","张朋","申振华"],"publication_date":"2026-02-24","filing_date":"2026-01-26","priority_date":"2026-01-26","cpc_codes":["G","G05","G05B","G05B13/00","G05B13/02","G05B13/04","G05B13/042","E","E21","E21B","E21B43/00","E21B43/16","E21B43/164","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N5/00","G06N5/04","G06N5/046","Y","Y02","Y02P","Y02P90/00","Y02P90/70"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121559900A/en"},{"publication_number":"CN121560686A","title":"Observable method and system for unifying log indexes and tracking data","abstract":"The invention discloses an observable method and system for unifying log indexes and tracking data, wherein the method comprises the steps of collecting various types of original observed data, constructing a column type storage structure after unifying formatting, and carrying out layered storage; the method comprises the steps of optimizing data layout by using column-level metadata and a query mode feature library, implementing a multi-level index strategy, simultaneously establishing causal patterns among heterogeneous observed data through data cleaning, structured enhancement and context information association to generate the observed data with rich contexts, and finally constructing a high-performance query processing system based on a unified query engine and a sub-linear algorithm to realize progressive result return and visual analysis. The invention obviously improves the data retrieval efficiency and analysis depth, so that operation and maintenance personnel can more quickly position the root cause of the system problem, and the average time of fault diagnosis is effectively reduced.","assignee":"Beijing Yulore Innovation Technology Co ltd","inventors":["张世亮"],"publication_date":"2026-02-24","filing_date":"2026-01-26","priority_date":"2026-01-26","cpc_codes":["G","G06","G06F","G06F11/00","G06F11/30","G06F11/3065","G06F11/3072","G","G06","G06F","G06F11/00","G06F11/30","G06F11/3089","G06F11/3093","G","G06","G06F","G06F11/00","G06F11/30","G06F11/32","G06F11/324","G06F11/328","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/23","G06F18/232","G06F18/2321","G06F18/23213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2433","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/091","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06N","G06N7/00","G06N7/01"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121560686A/en"},{"publication_number":"CN121563653A","title":"Shopping clue studying and judging system and early warning method based on artificial intelligence","abstract":"本发明涉及购物研判技术领域，具体涉及一种基于人工智能的购物线索研判系统及预警方法，该预警方法包括：通过获取电商平台购物数据构建多渠道购物数据集，并生成合规购物数据；将合规购物数据与危险品名录数据库进行语义匹配，识别危险品购买记录，并对相关订单进行预处理，形成标准化订单数据；结合关联规则挖掘算法，识别危险品原材料组合购买模式，生成组合特征数据；通过对组合购买的时间间隔和购买频率进行时间序列分析，评估分散购买行为，得到商品关联分析结果；综合商品风险等级、购买行为异常特征及历史交易记录，计算风险评分，并与预设风险阈值比对，生成风险预警结果。本发明识别了潜在的危险品组合购买行为。 This invention relates to the field of shopping analysis technology, specifically to an artificial intelligence-based shopping clue analysis system and early warning method. The early warning method includes: constructing a multi-channel shopping dataset by acquiring shopping data from e-commerce platforms and generating compliant shopping data; semantically matching the compliant shopping data with a hazardous materials list database to identify hazardous materials purchase records and preprocessing related orders to form standardized order data; identifying hazardous materials raw material combination purchase patterns using association rule mining algorithms to generate combination feature data; evaluating dispersed purchasing behavior through time series analysis of the time interval and frequency of combination purchases to obtain product association analysis results; calculating a risk score based on the product risk level, abnormal purchasing behavior characteristics, and historical transaction records, and comparing it with a preset risk threshold to generate a risk warning result. This invention identifies potential hazardous materials combination purchase behaviors.","assignee":"Beijing Qijun Technology Co ltd","inventors":["任亮"],"publication_date":"2026-02-24","filing_date":"2026-01-26","priority_date":"2026-01-26","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0607","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/26","G","G06","G06F","G06F21/00","G06F21/60","G06F21/602","G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6245","G06F21/6254","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/289","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0633","G06Q30/06331","G","G06","G06F","G06F2123/00","G06F2123/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121563653A/en"},{"publication_number":"CN121561385A","title":"Automatic feature engineering method based on Monte Carlo tree search and large language model","abstract":"本发明公开了基于蒙特卡洛树搜索与大语言模型的自动特征工程方法，包括向计算机系统输入表格数据集等特征数据；计算机系统执行特征初始化，评估输入的每个初始特征的得分，并构建初始特征树结构；迭代执行候选特征选择、基于大语言模型的扩展、特征得分评估及得分参数更新步骤，直至满足终止条件；基于选定特征，结合提示词工程调用大语言模型，扩展生成新特征节点；将新特征节点代入预定义的端到端机器学习pipeline评估，综合计算特征得分；沿特征合成路径自底向上更新特征树中所有祖先节点的相关参数；终止迭代后，输出能提升基础机器学习模型预测性能的最优特征集。本发明可以显著提升表格数据预测任务中机器学习模型的准确率。 This invention discloses an automatic feature engineering method based on Monte Carlo tree search and a large language model. The method includes inputting feature data, such as tabular datasets, into a computer system; the computer system performing feature initialization, evaluating the score of each initial feature, and constructing an initial feature tree structure; iteratively executing candidate feature selection, large language model-based expansion, feature score evaluation, and score parameter update steps until a termination condition is met; based on the selected features, combining prompt word engineering with the large language model to expand and generate new feature nodes; substituting the new feature nodes into a predefined end-to-end machine learning pipeline for evaluation, and comprehensively calculating feature scores; updating the relevant parameters of all ancestor nodes in the feature tree from bottom to top along the feature synthesis path; and after terminating the iteration, outputting the optimal feature set that improves the prediction performance of the basic machine learning model. This invention can significantly improve the accuracy of machine learning models in tabular data prediction tasks.","assignee":"Guizhou Youlian Borui Technology Co ltd; Guizhou University","inventors":["李晖","覃国忠","许玉田","陈攀峰","闵圣天","夏圣杰"],"publication_date":"2026-02-24","filing_date":"2026-01-26","priority_date":"2026-01-26","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/211","G06F18/2115","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06F","G06F18/00","G06F18/20","G06F18/27","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N7/00","G06N7/01"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121561385A/en"},{"publication_number":"KR20260022998A","title":"Method, program and apparatus for demand prediction based on machine learning","abstract":"본 개시는 머신러닝 기반의 수요 예측 방법 및 그 장치에 관한 것으로서, 컴퓨팅 장치에 의해 수행되는, 머신러닝을 기반으로 수요를 예측하는 방법으로서, 사용자 입력을 기초로 설정된 수요 예측 대상에 대한 과거 수요 데이터, 사용자 입력을 기초로 설정된 타겟 변수 및 사용자 입력을 기초로 설정된 작업 옵션 데이터를 포함한 입력 데이터를 준비하는 단계; 상기 과거 수요 데이터에 존재하는 타겟 변수를 확인하고, 머신러닝 기반의 제1 모델을 이용하여, 상기 과거 수요 데이터를 군집화 하는 단계; 머신러닝 기반의 제2 모델을 기초로, 상기 군집화를 통해 생성된 클러스터들 각각에서 유효 변수를 선택하는 단계; 및 상기 선택된 유효 변수와 상기 작업 옵션 데이터를 기반으로, 상기 타겟 변수에 대한 예측을 수행하는 머신러닝 기반의 제3 모델을 구축하는 단계;을 제공하고자 한다. The present disclosure relates to a machine learning-based demand prediction method and a device therefor, and provides a method for predicting demand based on machine learning, which is performed by a computing device, comprising: a step of preparing input data including past demand data for a demand prediction target set based on user input, a target variable set based on user input, and work option data set based on user input; a step of identifying a target variable present in the past demand data, and clustering the past demand data using a machine learning-based first model; a step of selecting a valid variable from each of the clusters generated through the clustering based on a machine learning-based second model; and a step of constructing a machine learning-based third model that performs a prediction on the target variable based on the selected valid variable and the work option data.","assignee":"(주)임팩티브에이아이","inventors":["정두희","정희원"],"publication_date":"2026-02-20","filing_date":"2026-02-10","priority_date":"2023-04-06","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0202","G","G06","G06F","G06F18/00","G06F18/20","G06F18/23","G06F18/232","G06F18/2321","G06F18/23213","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260022998A/en"},{"publication_number":"KR20260023546A","title":"Method and apparatus for providing design work platform","abstract":"본 개시의 실시예에 따라, 컴퓨팅 장치에 의해 수행되는, 디자인 작업 플랫폼에서의 디자인 작업 요청을 지원하는 방법은, 클라이언트 장치로부터, 디자인 작업을 요청하기 위한 작업 요청 데이터를 수신하는 단계, 상기 수신된 작업 요청 데이터가 작업 비효율성 요인을 포함하는지를 결정하는 단계 및 상기 수신된 작업 요청 데이터가 상기 작업 비효율성 요인을 포함하면 상기 작업 요청 데이터를 재획득하도록 상기 클라이언트 장치로 제안하는 단계를 포함할 수 있다. According to an embodiment of the present disclosure, a method for supporting a design task request in a design task platform, performed by a computing device, may include the steps of: receiving task request data for requesting a design task from a client device; determining whether the received task request data includes a task inefficiency factor; and suggesting to the client device to reacquire the task request data if the received task request data includes the task inefficiency factor.","assignee":"주식회사 테이아","inventors":["김서진","이보형"],"publication_date":"2026-02-20","filing_date":"2026-02-03","priority_date":"2022-11-11","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/103","G","G06","G06F","G06F40/00","G06F40/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G","G06","G06Q","G06Q50/00","G06Q50/50"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260023546A/en"},{"publication_number":"KR20260023987A","title":"System and Method for Integrated Standard Generation Based on Automatic Conflict Adjustment of Multiple Criteria","abstract":"본 발명은 복수의 질환 관리 기준, 산업 계측 기준, 규제 기준 및 운영 정책 기준 간 충돌을 정량적으로 분석하고 자동 조정하여 신규 통합 기준을 실시간 생성하는 시스템 및 방법에 관한 것이다. 본 발명은 기준 수집, 파라미터 변환, 충돌 강도 분석, 목적함수 기반 자동 조정 및 통합 기준 생성을 포함함으로써 기준 충돌 문제를 근본적으로 해소한다. 이에 따라 분석 정확성과 시스템 신뢰성이 향상되며, 의료, 산업 안전, 에너지 관리 및 환경 관리 분야 전반에 적용 가능하다. The present invention relates to a system and method for quantitatively analyzing and automatically adjusting conflicts among multiple disease management standards, industrial measurement standards, regulatory standards, and operational policy standards to generate new, integrated standards in real time. The present invention fundamentally resolves the issue of standard conflict by encompassing standard collection, parameter transformation, conflict intensity analysis, objective function-based automatic adjustment, and integrated standard generation. This enhances analysis accuracy and system reliability, and is applicable across the fields of healthcare, industrial safety, energy management, and environmental management.","assignee":"구교선; 구현우","inventors":["구교선","구현우"],"publication_date":"2026-02-20","filing_date":"2026-02-02","priority_date":"2026-02-02","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06F","G06F17/00","G06F17/10","G06F17/18","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06316","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0633","G","G06","G06Q","G06Q10/00","G06Q10/10"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260023987A/en"},{"publication_number":"KR20260022985A","title":"Image depth estimation methods, systems, electronic devices, and media","abstract":"본 개시는, 이미지 깊이 추정 방법, 시스템, 전자 장치 및 매체에 관한다. 포인트 클라우드 수집 장치가 목적 대상에 대해 수집한 다중 프레임의 포인트 클라우드와 이미지 수집 장치가 목적 대상에 대해 수집한 적어도 한 장의 이미지를 획득하고, 포인트 클라우드 수집 장치의 외부 파라미터에 기반하여 다중 프레임의 포인트 클라우드 중의 각 프레임의 포인트 클라우드를 수집한 제1 타임 스탬프에 대응하는 제1 위치자세를 결정하고,제1 위치자세에 기반하여 다중 프레임의 포인트 클라우드를 밀집화하여 밀집 포인트 클라우드를 얻고, 적어도 두 프레임의 포인트 클라우드의 제1 타임 스탬프에 대응하는 제1 위치자세에 기반하여 이미지를 수집한 제2 타임 스탬프에 대응하는 제2 위치자세를 결정하고, 이미지 수집 장치의 내부 파라미터 및 제2 위치자세에 기반하여 밀집 포인트 클라우드를 이미지에 투영하여 이미지 중 픽셀의 깊이 정보를 결정한다. The present disclosure relates to an image depth estimation method, a system, an electronic device, and a medium. A point cloud collection device acquires a point cloud of multiple frames collected for a target object and at least one image collected for a target object by an image collection device, determines a first position and attitude corresponding to a first time stamp of collecting a point cloud of each frame among the point clouds of the multiple frames based on an external parameter of the point cloud collection device, obtains a dense point cloud by densifying the point clouds of the multiple frames based on the first position and attitude, determines a second position and attitude corresponding to a second time stamp of collecting an image based on the first position and attitude corresponding to the first time stamp of the point clouds of at least two frames, and projects the dense point cloud onto an image based on an internal parameter and the second position and attitude of the image collection device to determine depth information of a pixel in the image.","assignee":"선전 디-로보틱스 씨오., 엘티디.","inventors":["씬루이 멍"],"publication_date":"2026-02-20","filing_date":"2026-01-27","priority_date":"2025-09-30","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/50","G06T7/55","G","G06","G06N","G06N20/00","G","G06","G06T","G06T3/00","G06T3/06","G","G06","G06T","G06T7/00","G06T7/70","G06T7/73","G","G06","G06T","G06T7/00","G06T7/80","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10028"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260022985A/en"},{"publication_number":"AU2026200659A1","title":"Intelligently generating and managing third-party sources within a contextual hub","abstract":"1006373109 The present disclosure is directed toward systems, methods, and non-transitory computer readable media for generating contextual hubs for organizing and presenting web- accessible content from third-party sources. In particular, the systems described herein can organize and manage within a contextual hub. For instance, the disclosed systems may perform actions on tabs based on analyzing usage signals associated with the tabs. Furthermore, the disclosed systems can organize contextually related content within contextual hubs. The disclosed systems may also facilitate collaboration between users within a contextual hub by synchronizing interactions with content within a contextual hub.","assignee":"Dropbox Inc","inventors":["Hudson Arnold","Thomas Kleinpeter","Terrence Mcardle","Kristoffer Mendoza"],"publication_date":"2026-02-19","filing_date":"2026-01-30","priority_date":"2020-06-08","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/954","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/955","G06F16/9558","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/955","G06F16/9562","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/957","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/958","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481","G06F3/0483","G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/451","G","G06","G06N","G06N20/00","G","G06","G06N","G06N5/00","G06N5/04","H","H04","H04L","H04L67/00","H04L67/01","H04L67/02","H","H04","H04L","H04L67/00","H04L67/01","H04L67/12","H04L67/125","H","H04","H04L","H04L67/00","H04L67/50","H04L67/535","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026200659A1/en"},{"publication_number":"AU2026200721A1","title":"Travel-specific natural language processing system","abstract":"Systems and methods for travel-specific natural language processing are disclosed. The techniques described herein include obtaining input text data having a noisy text format and corresponding to a travel-specific lexicon. The techniques include retrieving additional text data having the noisy text format from one or more data sources, the additional text data corresponding to the travel-specific lexicon, and causing a machine-learning model to generate text output based on (i) the input text data and (ii) the additional text data retrieved from the one or more data sources.","assignee":"Expedia Inc","inventors":["Mani NAJMABADI","Niloofar Safi SAMGHABADI"],"publication_date":"2026-02-19","filing_date":"2026-01-30","priority_date":"2023-02-24","cpc_codes":["G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06F","G06F40/00","G06F40/20","G","G06","G06F","G06F40/00","G06F40/40","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026200721A1/en"},{"publication_number":"KR20260022353A","title":"Apparatus for providing investing information on digital assets","abstract":"디지털 자산 투자정보 제공장치가 개시된다. 본 발명은 디지털 자산 데이터를 수집하고 가공 및 표준화를 포함한 분석 과정을 거쳐 디지털 자산 투자정보를 생성하여 이를 제공한다. A digital asset investment information provider is disclosed. The present invention collects digital asset data, processes it through an analysis process including processing and standardization, and then generates and provides digital asset investment information.","assignee":"주식회사 보난자랩","inventors":["박혜연","전효연"],"publication_date":"2026-02-19","filing_date":"2026-01-27","priority_date":"2024-03-12","cpc_codes":["G","G06","G06Q","G06Q40/00","G06Q40/04","G06Q40/042","G06Q40/0421","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q40/00","G06Q40/06","G06Q40/063","G06Q40/0631","H","H04","H04L","H04L9/00","H04L9/50"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260022353A/en"},{"publication_number":"AU2026200561A1","title":"Systems and methods for deriving and optimizing classifiers from multiple datasets","abstract":"70 Abstract: Systems and methods for subject clinical condition evaluation using a plurality of modules are provided. Modules comprise features whose corresponding feature values associate with an absence, presence or stage of phenotypes associated with the clinical condition. A first dataset is obtained having feature values, acquired through a first technical background from respective subjects in transcriptomic, proteomic, or metabolomic form, for at least a first of the plurality of modules. A second training dataset is obtained having feature values, acquired through a technical background other than the first technical background, from training subjects of the second dataset, in the same form as for the first dataset, of at least the first module. Inter-dataset batch effects are removed by co- normalizing feature values across the training datasets, thereby calculating co-normalized feature values used to train a classifier for clinical condition evaluation of the test subject. 70 20 26 20 05 61 27 J an 2 02 6 2 0 2 6 2 0 0 5 6 1 2 7 J a n 2 0 2 6 7 0","assignee":"Inflammatix Inc","inventors":["Ljubomir BUTUROVIC","Purvesh Khatri","Roland Luethy","Michael B. Mayhew","Timothy E. Sweeney"],"publication_date":"2026-02-19","filing_date":"2026-01-27","priority_date":"2019-03-22","cpc_codes":["G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N7/00","G06N7/01","G","G16","G16B","G16B20/00","G","G16","G16B","G16B25/00","G16B25/10","G","G16","G16B","G16B40/00","G","G16","G16B","G16B40/00","G16B40/20","G","G16","G16B","G16B40/00","G16B40/30","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/30"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026200561A1/en"},{"publication_number":"AU2026200500A1","title":"Firearm monitoring and remote support system","abstract":"A system for firearm monitoring and remote support comprises a plurality of firearms within a deployment location, wherein each firearm includes one or more sensors that record sensor information used to produce a signal; response infrastructure configured for deployment to the deployment location; and a server device running application software that uses the signals received from each of the firearms to detect a threat within the deployment location and causes the deployment of the response infrastructure to the deployment location, wherein the response infrastructure supports users of the plurality of firearms in addressing the detected threat. WO 2020/077254 PCT/US2019/055925 1/37 104 106 108 WEARABLE STATIONARY FIREARMS DEVICES DEVICES SENSORS 118 SENSORS 120 SENSORS 122 CONNECTION POINT 116 110 114 RESPONSE NETWORK INFRASTRUCTURE 102 APPLICATION 112 DATABASE 124 SERVER DEVICE FIG. 1 20 26 20 05 00 23 J an 2 02 6 W O 2 0 2 0 / 0 7 7 2 5 4 P C T / U S 2 0 1 9 / 0 5 5 9 2 5 1 / 3 7 2 0 2 6 2 0 0 5 0 0 2 3 J a n 2 0 2 6 1 0 6 1 0 8 W E A R A B L E S T A T I O N A R Y F I R E A R M S S E N S O R S 1 1 8 S E N S O R S 1 2 0 S E N S O R S 1 2 2 C O N N E C T I O N 1 1 6 1 1 0 1 1 4 R E S P O N S E N E T W O R K 1 0 2 A P P L I C A T I O N 1 1 2 D A T A B A S E 1 2 4 S E R V E R D E V I C E F I G . 1","assignee":"Armaments Research Co Inc","inventors":["Michael Canty","William DENG"],"publication_date":"2026-02-19","filing_date":"2026-01-23","priority_date":"2018-10-12","cpc_codes":["H","H04","H04L","H04L67/00","H04L67/01","H04L67/12","F","F41","F41G","F41G3/00","F41G3/14","F41G3/147","G","G06","G06F","G06F1/00","G06F1/16","G06F1/1613","G06F1/163","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481","G06F3/04817","G","G06","G06N","G06N20/00","G","G06","G06T","G06T15/00","G06T15/005","G","G06","G06V","G06V20/00","G06V20/20","H","H04","H04L","H04L67/00","H04L67/50","H04L67/52","H","H04","H04W","H04W4/00","H04W4/02","H","H04","H04W","H04W4/00","H04W4/02","H04W4/025","H04W4/026","H","H04","H04W","H04W4/00","H04W4/02","H04W4/025","H04W4/027","H","H04","H04W","H04W4/00","H04W4/30","H04W4/38","F","F41","F41G","F41G3/00","F41G3/04","F","F41","F41G","F41G9/00","G","G06","G06V","G06V20/00","G06V20/10","G06V20/13"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026200500A1/en"},{"publication_number":"DE202026100532U1","title":"Deep learning to improve dose reconstruction for adaptive radiotherapy in real time","abstract":"Ein computerimplementiertes System zur Verbesserung der Strahlendosisrekonstruktion in der adaptiven Echtzeit-Strahlentherapie, wobei das System ein Datenerfassungsmodul zur Erfassung von Behandlungsdaten in Echtzeit während der Strahlentherapie, ein Vorverarbeitungsmodul zur Vorverarbeitung der erfassten Daten, eine Deep-Learning-Dosisrekonstruktions-Engine zur Rekonstruktion der abgegebenen Strahlendosisverteilung mithilfe eines trainierten Deep-Learning-Modells, ein Dosisbewertungs- und Vergleichsmodul zum Vergleich der rekonstruierten Dosisverteilung mit einer geplanten Dosisverteilung und ein adaptives Entscheidungsunterstützungsmodul zur Unterstützung der Anpassung der Strahlentherapie-Behandlungsparameter auf der Grundlage der Vergleichsergebnisse umfasst. A computer-implemented system for improving radiation dose reconstruction in adaptive real-time radiotherapy, comprising a data acquisition module for capturing treatment data in real time during radiotherapy, a preprocessing module for preprocessing the captured data, a deep-learning dose reconstruction engine for reconstructing the delivered radiation dose distribution using a trained deep-learning model, a dose evaluation and comparison module for comparing the reconstructed dose distribution with a planned dose distribution, and an adaptive decision support module to assist in adjusting radiotherapy treatment parameters based on the comparison results.","assignee":"Individual","inventors":[],"publication_date":"2026-02-18","filing_date":"2026-02-01","priority_date":"2026-02-01","cpc_codes":["G","G16","G16H","G16H30/00","G16H30/20","A","A61","A61N","A61N5/00","A61N5/10","A61N5/1048","A61N5/1064","A61N5/1065","A61N5/1067","G","G06","G06N","G06N3/00","G06N3/02","G","G16","G16H","G16H15/00","G","G16","G16H","G16H20/00","G16H20/40","G","G16","G16H","G16H30/00","G16H30/40","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/70"],"country":"DE","kind":"application","source_url":"https://patents.google.com/patent/DE202026100532U1/en"},{"publication_number":"CN121544087A","title":"A method for optimizing the operation of hydrogen-containing building energy systems using a multi-role large model","abstract":"The invention discloses a multi-role large-model assisted hydrogen-containing building energy system operation optimization method, which belongs to the technical field of building energy system optimization control and comprises the steps of firstly establishing a problem of minimizing the operation cost of a hydrogen-containing building multi-energy system in an off-grid operation mode, secondly re-modeling the problem into a safe Markov decision process, defining a system state space, an action space and a composite rewarding function, then solving the modeled safe Markov decision process based on a multi-role large-language model assisted near-end strategy optimization algorithm to obtain an agent operation strategy related to the hydrogen-containing building multi-energy system, and finally, carrying out online decision based on the obtained optimization strategy by an agent and acting the decision on an actual hydrogen-containing building multi-energy system.","assignee":"Nanjing University of Posts and Telecommunications","inventors":["余亮","胡红伟","方景","陈志强","岳东","张廷军"],"publication_date":"2026-02-17","filing_date":"2026-01-22","priority_date":"2026-01-22","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G","G06","G06N","G06N3/00","G06N3/004","G06N3/008","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06Q","G06Q50/00","G06Q50/06","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/16","G06Q50/163","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121544087A/en"},{"publication_number":"CN121542311A","title":"A Data Weight Quantization Method and System Based on Operator-Level Lineage Analysis","abstract":"本申请公开了一种基于算子级血缘解析的数据权重量化方法及系统，该方法将目标场景对应的待解析SQL语句解析为抽象语法树；遍历抽象语法树，提取多种算子，并构建多种算子之间的依赖关系；基于多种算子和依赖关系，构建算子有向无环图；基于算子有向无环图，构建包含算子节点、表节点以及节点之间的关联关系的算子级血缘存储模型，关联关系的属性包含边权重值，边权重值通过数据流量权重、算子复杂度权重、依赖关系权重以及路径深度权重加权求和得到；将边权重值进行归一化，得到归一化后的边权重值，以实现数据权重量化。本申请能够为数据治理中的数据质量溯源和查询优化中的关键算子定位，提供细粒度和可量化的血缘分析支撑。 This application discloses a data weight quantification method and system based on operator-level lineage analysis. The method parses the SQL statement to be parsed for the target scenario into an abstract syntax tree; traverses the abstract syntax tree, extracts various operators, and constructs dependencies between these operators; based on the operators and dependencies, constructs an operator-directed acyclic graph (DAG); based on the DAG, constructs an operator-level lineage storage model containing operator nodes, table nodes, and relationships between nodes. The attributes of these relationships include edge weights, which are obtained by weighted summation of data flow weights, operator complexity weights, dependency weights, and path depth weights; and normalizes the edge weights to obtain normalized edge weights, thus achieving data weight quantification. This application can provide fine-grained and quantifiable lineage analysis support for data quality tracing in data governance and key operator location in query optimization.","assignee":"National University of Defense Technology","inventors":["任小丽","李小勇","邵成成","朱湘","任开军","邓科峰","林家润","谭家明","陈信宇","王雅真"],"publication_date":"2026-02-17","filing_date":"2026-01-22","priority_date":"2026-01-22","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2455","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/242","G06F16/2433","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2452","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","Y","Y02","Y02D","Y02D10/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121542311A/en"},{"publication_number":"CN121542914A","title":"A Health Status Monitoring Method for Intelligent EMB Systems Based on Sensor and Control Integration","abstract":"The invention relates to the technical field of EMB system detection data processing, in particular to an intelligent EMB system health state monitoring method based on sensing and control integration, which comprises the steps of collecting detection data generated by the operation of an electromechanical braking system in the braking process of a vehicle; the detection data comprise motor monitoring data, single comprehensive loss of the electromechanical braking system in each braking process is calculated according to the motor monitoring data, a window abnormality index corresponding to a sliding window is obtained according to the single comprehensive loss, a data characteristic space is built by combining the single comprehensive loss, the data characteristic space is processed by adopting an isolated forest algorithm to obtain isolated factors of each braking, and health status classification is carried out on the obtained isolated factors according to a preset health status grade threshold. The invention can remarkably improve the detection sensitivity and the robustness of early abrasion and micro faults.","assignee":"Hubei Yukong Zhiqu Technology Co ltd","inventors":["张涛","倪春阳"],"publication_date":"2026-02-17","filing_date":"2026-01-22","priority_date":"2026-01-22","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/24323","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06N","G06N5/00","G06N5/01"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121542914A/en"},{"publication_number":"CN121543047A","title":"A Missing Multivariate Time Series Prediction Method Integrating Spatiotemporal Features and Frequency Domain Information","abstract":"The invention discloses a missing multiple time sequence prediction method for fusing space-time characteristics and frequency domain information, which relates to the field of time sequence prediction and comprises the following steps of obtaining multiple time sequence data containing missing values and a corresponding missing matrix; the method comprises the steps of inputting the time characteristics into a time characteristic reconstruction module, interpolating missing data through an attention weight decline mechanism to obtain reconstructed time characteristics, utilizing a space-time characteristic extraction module to extract spatial correlation characteristics among variables through transposition and a multi-head self-attention mechanism to obtain space-time fusion characteristics, conducting fusion of time domain information and frequency domain information through a plurality of cascaded frequency domain enhancement units through a frequency domain enhancement module, outputting the enhanced time sequence characteristics, and inputting the time sequence characteristics into a prediction output layer to obtain a predicted value of a future time step. The method realizes end-to-end accurate prediction of missing multi-element time sequence data, and improves robustness and periodic feature capturing capability of the model under the condition of data missing.","assignee":"Zhejiang Normal University CJNU","inventors":["陈丽娜","章峻波","高宏"],"publication_date":"2026-02-17","filing_date":"2026-01-22","priority_date":"2026-01-22","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2131","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121543047A/en"},{"publication_number":"CN121543581A","title":"A cross-document semantic consistency verification system for bidding and tendering scenarios","abstract":"本发明公开了一种面向招投标场景的跨文档语义一致性校验系统及方法，涉及数据处理技术领域，包括：采集招投标文档数据并进行预处理；分析招投标文档数据中语义单元的关键程度构建术语关键指数；基于术语关键指数分析招投标文档的关键语义单元的风险程度并构建锚点风险熵；基于锚点风险熵分析所述招投标文档数据和对比文档数据的公共关键语义单元的风险敏感程度并构建跨文档风险因子；基于所述跨文档风险因子改进孪生神经网络的对比损失函数，实现对不同招投标文档数据的语义一致性校验。本发明解决了模型在训练过程中对关键不一致性与非关键不一致性的区分度较低，从而影响了系统对高风险错误的识别精度的问题。 This invention discloses a cross-document semantic consistency verification system and method for bidding and tendering scenarios, relating to the field of data processing technology. The system includes: collecting and preprocessing bidding and tendering document data; analyzing the criticality of semantic units in the bidding and tendering document data to construct a terminology key index; analyzing the risk level of key semantic units in the bidding and tendering documents based on the terminology key index and constructing an anchor point risk entropy; analyzing the risk sensitivity of common key semantic units in the bidding and tendering document data and comparative document data based on the anchor point risk entropy and constructing a cross-document risk factor; and improving the contrastive loss function of the Siamese neural network based on the cross-document risk factor to achieve semantic consistency verification of different bidding and tendering document data. This invention solves the problem that the model has low discrimination between key inconsistencies and non-key inconsistencies during training, thus affecting the system's accuracy in identifying high-risk errors.","assignee":"Jiangsu Share Sun Information Technology Co ltd","inventors":["朱慧昌","赵明光","王鹏程","何思龙","高扬","王建房"],"publication_date":"2026-02-17","filing_date":"2026-01-22","priority_date":"2026-01-22","cpc_codes":["G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G06F40/226","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G06F40/216","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121543581A/en"},{"publication_number":"CN121543472A","title":"A method for predicting the cost-effectiveness of solid rocket propulsion based on coordinate descent depth integration","abstract":"本发明涉及航空航天飞行器总体设计等领域，公开了一种基于坐标下降深度集成的固体火箭动力成本效能预测方法，包括：构建固体火箭发动机的多源融合数据库，并训练基于坐标下降优化的集成稀疏神经网络；所述集成稀疏神经网络包含加性模型，加性模型由基学习器和对应的权重系数进行构造；在集成稀疏神经网络的各个基学习器的训练中，构建包含均方误差项和 范数正则化项的目标函数；所述 范数正则化项用于对基学习器的输入层到隐层的权重矩阵施加行稀疏约束；针对固体火箭发动机的新设计方案，获取其特征参数并构造输入特征，将其输入到训练好的集成稀疏神经网络，得到预测的输出响应，用于对固体火箭发动机动力系统进行成本估算或效能估算；本发明收敛速度快且精度高。 This invention relates to the fields of aerospace vehicle overall design, and discloses a method for predicting the cost-effectiveness of solid rocket propulsion based on coordinate descent depth integration. The method includes: constructing a multi-source fusion database for solid rocket engines and training an integrated sparse neural network based on coordinate descent optimization; the integrated sparse neural network includes an additive model, which is constructed from base learners and corresponding weight coefficients; in the training of each base learner of the integrated sparse neural network, a method is constructed that includes a mean squared error term and... The objective function of the norm regularization term; The norm regularization term is used to apply sparse constraints to the weight matrix from the input layer to the hidden layer of the base learner; for a new design scheme of solid rocket motor, its feature parameters are obtained and input features are constructed, which are then input into a trained ensemble sparse neural network to obtain the predicted output response, which is used to estimate the cost or efficiency of the solid rocket motor propulsion system; this invention has fast convergence speed and high accuracy.","assignee":"Xian Institute of Modern Control Technology","inventors":["刘钧圣","骆盛","李国旭","许琛","乔浩","高登巍"],"publication_date":"2026-02-17","filing_date":"2026-01-22","priority_date":"2026-01-22","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06F","G06F16/00","G06F16/20","G06F16/21","G06F16/211","G06F16/212","G","G06","G06F","G06F30/00","G06F30/10","G06F30/15","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06F","G06F2111/00","G06F2111/04","G","G06","G06F","G06F2111/00","G06F2111/06"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121543472A/en"},{"publication_number":"CN121543749A","title":"A Large-Model-Based Intelligent Question-Answering Management System and Method for Chinese Language","abstract":"The invention discloses a language intelligent question-answering management system and method based on a large model, which are characterized in that real-time language question data of a user are obtained, chinese texts are segmented by utilizing recursive character segmentation, a knowledge base is constructed by combining automatic information extraction, and the knowledge base is updated in real time by utilizing incremental learning to obtain a target knowledge base; the method comprises the steps of obtaining historical dialogue data, identifying user intention through a model to generate an initial reply strategy, converting the real-time language question data into vector representation by using a pre-trained RoBERTa semantic embedding model, calculating the semantic similarity of user questions and candidate answers through a GNN graph neural network by combining entity relations of knowledge graphs in a target knowledge base, and generating the target reply strategy based on the initial reply strategy and the semantic similarity. And the accuracy and efficiency of question and answer are improved.","assignee":"Guizhou Yuhao Technology Development Co ltd","inventors":["杨见会","连娟娟","平艳丽","宋彩锋","韩露","陈金明"],"publication_date":"2026-02-17","filing_date":"2026-01-22","priority_date":"2026-01-22","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06F","G06F16/00","G06F16/20","G06F16/22","G06F16/2228","G06F16/2237","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/242","G06F16/243","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2455","G06F16/24564","G","G06","G06F","G06F16/00","G06F16/20","G06F16/28","G06F16/284","G06F16/285","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/289","G06F40/295","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121543749A/en"},{"publication_number":"CN121541148A","title":"A Weather Radar Clutter Identification Method and System Based on Fuzzy Reasoning","abstract":"The invention discloses a weather radar wind turbine clutter recognition method and system based on fuzzy reasoning, which belong to the technical field of fuzzy reasoning, wherein a monitoring area of a target weather radar is divided into a plurality of echo observation subunits according to a range gate and a scanning azimuth angle, and the reflectivity factors, radial speed, spectral width and differential reflectivity of the echo observation subunits are continuously acquired in a preset data acquisition time sequence. The method has the advantages that the echo stability characteristic quantity is constructed by analyzing the discrete characteristics of the reflectivity factors and the radial speed on the time sequence, the echo form consistency characteristic quantity is constructed based on the consistency deviation of the spectral width and the differential reflectivity in the adjacent domain, the echo stability characteristic quantity and the echo form consistency characteristic quantity are subjected to fuzzy mapping, the distinguishing and the identification of wind turbine clutter and meteorological echoes are realized through fuzzy comprehensive judgment quantity, the accuracy and the robustness of wind turbine clutter identification are improved, and the method is suitable for weather radar data quality control under the complex background condition.","assignee":"Nanjing Institute Of Meteorological Science And Technology Innovation; Jiangsu Meteorological Observation Center Jiangsu Jintan Meteorological Comprehensive Test Base","inventors":["刘寅","曾强宇","苏明月","张逸扬","高尧","刘端阳","王宏斌","祖繁"],"publication_date":"2026-02-17","filing_date":"2026-01-22","priority_date":"2026-01-22","cpc_codes":["G","G01","G01S","G01S7/00","G01S7/02","G01S7/021","G","G01","G01S","G01S13/00","G01S13/88","G01S13/95","G","G01","G01S","G01S7/00","G01S7/02","G01S7/41","G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06N","G06N5/00","G06N5/04","G06N5/048","Y","Y02","Y02A","Y02A90/00","Y02A90/10"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121541148A/en"},{"publication_number":"CN121545615A","title":"A rapid detection method for complex enzyme activity integrating electrochemical sensing and artificial intelligence","abstract":"The invention discloses a method for rapidly detecting the activity of complex enzyme by fusing electrochemical sensing and artificial intelligence, and particularly relates to the technical field of electrochemical detection and signal analysis; the method is used for solving the problem that the contribution and the synergistic effect of each component are difficult to analyze because the existing detection method adopting fixed potential excitation cannot be simultaneously adapted to the optimal response conditions of each component in the complex enzyme; the method is realized by applying a preset composite potential excitation sequence to an electrochemical sensing system to synchronously excite each enzyme component to respond, collecting multidimensional electrochemical response signals, reconstructing the multidimensional electrochemical response signals into phase space tracks, further extracting recursion quantitative characteristics and Poincare cross-section point set distribution characteristics, judging the system dynamics state based on the characteristics and adaptively combining the characteristics to form a global electrochemical characteristic vector, analyzing contribution degree information of different catalytic behavior modes through a pre-trained artificial intelligent model, and finally generating a comprehensive evaluation result comprising an overall efficiency index and an efficiency grade by weighting and fusing the contribution degree information.","assignee":"Friendly Pharmaceutical","inventors":["叶长蓉","龚征强"],"publication_date":"2026-02-17","filing_date":"2026-01-22","priority_date":"2026-01-22","cpc_codes":["G","G16","G16C","G16C20/00","G16C20/20","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06F","G06F18/00","G06F18/20","G06F18/27","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G16","G16C","G16C20/00","G16C20/70"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121545615A/en"},{"publication_number":"CO2026001219A2","title":"Systems and methods for developing a knowledge base composed of data collected from countless sources","abstract":"Se proporciona un sistema para construir una base de conocimientos. El sistema incluye un procesador. El sistema incluye además un medio de almacenamiento legible por computadora no transitorio que contiene instrucciones las cuales, cuando se ejecutan en el procesador, hacen que el sistema realice operaciones. Las operaciones incluyen recibir datos multimodales de una o más fuentes; analizar los datos para determinar características; almacenar las características en una base de datos; recibir una consulta de búsqueda para buscar las características almacenadas; analizar la consulta de búsqueda utilizando un gran modelo de lenguaje para extraer características de búsqueda; generar resultados de búsqueda a partir de las características de búsqueda; y mostrar los resultados de búsqueda en una interfaz gráfica de usuario estandarizada que incluye una leyenda que tiene al menos una o más de las características de búsqueda mostradas. A system is provided for building a knowledge base. The system includes a processor. The system also includes a non-transient, machine-readable storage medium containing instructions which, when executed by the processor, cause the system to perform operations. These operations include receiving multimodal data from one or more sources; analyzing the data to determine features; storing the features in a database; receiving a search query to retrieve the stored features; analyzing the search query using a large model language to extract search features; generating search results from the search features; and displaying the search results in a standardized graphical user interface that includes a legend containing at least one or more of the displayed search features.","assignee":"Red Atlas Inc","inventors":["Ortega Oscar David Corredor","Henry Forsyth Keenan","Duque Andrés Felipe Valencia","Castillo Juan David Martínez","Rosales Alejandro Dominguez","Buriticá Andrés Pérez","Jose Martinez"],"publication_date":"2026-02-13","filing_date":"2026-02-03","priority_date":"2023-07-03","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/20","G06F16/29","G","G06","G06F","G06F16/00","G06F16/20","G06F16/21","G06F16/215","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/242","G06F16/243","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2457","G06F16/24575","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/248","G","G06","G06F","G06F16/00","G06F16/20","G06F16/28","G06F16/284","G06F16/285","G06F16/287","G","G06","G06F","G06F16/00","G06F16/40","G06F16/43","G06F16/438","G","G06","G06N","G06N20/00","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04"],"country":"CO","kind":"application","source_url":"https://patents.google.com/patent/CO2026001219A2/en"},{"publication_number":"KR20260021672A","title":"Artificial refrigerator and method for controlling the same","abstract":"지능형 냉장고를 개시한다. 본 발명의 지능형 냉장고는 냉장고의 작동 상태를 감지하고, 상기 냉장고의 작동 상태에 대한 작동 정보를 획득하는 센싱부;와 상기 센싱부를 통해 획득된 상기 작동 정보에 기초하여 딥러닝 기반의 1차 진단 엔진을 통해 정상 또는 불량 여부를 판정하고, 불량 판정 시 딥러닝 기반의 2차 진단 엔진을 통해 불량 원인을 진단하는 프로세서;를 포함한다. 본 발명의 지능형 냉장고는, 사용자 단말기 및 서버 중 하나 이상이 인공 지능(Artificail Intelligenfce) 모듈, 드론(Unmanned Aerial Vehicle, UAV), 로봇, 증강 현실(Augmented Reality, AR) 장치, 가상 현실(virtual reality, VR) 장치, 5G 서비스와 관련된 장치 등과 연계될 수 있다. Disclosed is an intelligent refrigerator. The intelligent refrigerator of the present invention comprises: a sensing unit that detects the operating status of a refrigerator and obtains operating information about the operating status of the refrigerator; and a processor that determines whether the refrigerator is normal or defective through a deep learning-based primary diagnosis engine based on the operating information obtained through the sensing unit, and, if defective, diagnoses the cause of the defect through a deep learning-based secondary diagnosis engine. The intelligent refrigerator of the present invention may be connected to at least one of a user terminal and a server, such as an artificial intelligence (AI) module, an unmanned aerial vehicle (UAV), a robot, an augmented reality (AR) device, a virtual reality (VR) device, or a device related to a 5G service.","assignee":"엘지전자 주식회사","inventors":["한준수","양영훈","엄용환","정준성","한초록"],"publication_date":"2026-02-13","filing_date":"2026-01-21","priority_date":"2019-08-30","cpc_codes":["F","F25","F25B","F25B31/00","F25B31/006","F25B31/008","F","F25","F25B","F25B39/00","F25B39/04","F","F25","F25B","F25B49/00","F25B49/02","F","F25","F25B","F25B5/00","F25B5/02","F","F25","F25D","F25D11/00","F25D11/02","F25D11/022","F","F25","F25D","F25D29/00","F25D29/005","F","F25","F25D","F25D29/00","F25D29/006","F","F25","F25D","F25D29/00","F25D29/008","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06Q","G06Q50/00","G06Q50/10","H","H04","H04L","H04L12/00","H04L12/28","H04L12/2803","H04L12/2823","H04L12/2825","F","F25","F25B","F25B2500/00","F25B2500/06","F","F25","F25B","F25B2500/00","F25B2500/19","F","F25","F25B","F25B2600/00","F25B2600/25","F25B2600/2511","F","F25","F25B","F25B2700/00","F25B2700/02","F","F25","F25B","F25B2700/00","F25B2700/21","F25B2700/2104","F","F25","F25B","F25B2700/00","F25B2700/21","F25B2700/2106","F","F25","F25D","F25D2700/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260021672A/en"},{"publication_number":"CN121524177A","title":"A method for tracing the lineage and assessing the quality of cross-border e-commerce data","abstract":"The invention relates to the technical field of data management and quality monitoring, and discloses a cross-border electronic commerce data blood-margin tracking and quality assessment method. The method comprises the steps of establishing a metadata snapshot sequence of data nodes by constructing a global topological graph containing space and level attributes, and calculating node association strength according to the metadata snapshot sequence to generate a dynamic propagation path so as to reflect the real flowing state of the data blood edges in real time. The state characteristic vector of each node on the path is extracted, and the potential migration mode of the data quality state is identified by analyzing the evolution trend of the state characteristic vector. And fusing migration modes of all paths to construct a data quality evaluation matrix, and driving iterative optimization of a tracking strategy to form closed-loop monitoring. The method can dynamically capture the change of the blood-cause relationship, accurately position the source of the quality problem and forecast the propagation of the source of the quality problem, and improves the reliability and maintainability of a data system.","assignee":"Fujian Dongsiloh Technology Co ltd","inventors":["陈圣强"],"publication_date":"2026-02-13","filing_date":"2026-01-19","priority_date":"2026-01-19","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/21","G06F16/215","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2455","G06F16/24564","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2455","G06F16/24568","G","G06","G06F","G06F16/00","G06F16/20","G06F16/25","G06F16/254","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06395","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/083","G06Q10/0831","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/083","G06Q10/0838"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121524177A/en"},{"publication_number":"CN121528415A","title":"A reinforcement learning-based dynamic dosage control system for analgesics in critically ill patients","abstract":"本发明公开了基于强化学习的危重患者镇痛药物剂量动态调控系统，涉及强化学习技术领域。该系统包括：数据采集模块，用于获取目标危重患者的第一患者信息，其中，所述第一患者信息包括不变本征信息、病症类别信息与实时生理信息；状态映射模块，用于根据所述第一患者信息，结合强化学习方法构建状态映射模型，并输入所述实时生理信息至所述状态映射模型，获取目标危重患者的伪标准生理信息；调控决策模块，用于基于所述伪标准生理信息，根据预设的剂量调控规则生成针对镇痛药物的建议剂量指令。本发明有效提升了镇痛药物剂量的安全性与不同危重病症间的泛化能力。 This invention discloses a dynamic dosage control system for analgesics in critically ill patients based on reinforcement learning, belonging to the field of reinforcement learning technology. The system includes: a data acquisition module for acquiring first patient information of a target critically ill patient, wherein the first patient information includes invariant intrinsic information, disease category information, and real-time physiological information; a state mapping module for constructing a state mapping model based on the first patient information and using reinforcement learning methods, and inputting the real-time physiological information into the state mapping model to obtain pseudo-standard physiological information of the target critically ill patient; and a control decision module for generating recommended dosage instructions for analgesics based on the pseudo-standard physiological information and according to preset dosage control rules. This invention effectively improves the safety of analgesic drug dosage and its generalization ability across different critical illnesses.","assignee":"Nuclear Industry General Hospital","inventors":["毛自若","高稚淇"],"publication_date":"2026-02-13","filing_date":"2026-01-19","priority_date":"2026-01-19","cpc_codes":["G","G16","G16H","G16H20/00","G16H20/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H50/00","G16H50/20"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121528415A/en"},{"publication_number":"CN121527463A","title":"Clustering methods, apparatus, electronic devices and storage media for misaligned multi-view clustering","abstract":"The invention relates to a clustering method, a device, electronic equipment and a storage medium for unaligned multi-views, which comprise the steps of constructing an adjacent graph for each view, matching a graph structure of the adjacent graph of a non-reference view with the adjacent graph of a reference view through a replacement matrix, inputting each view and the adjacent graph thereof into a first layer graph rolling network, extracting depth representation of each view, constructing a local adjacent graph of each view based on the adjacent graph and the depth representation of each view, updating the neighbor number of the local adjacent graph at intervals of at least one training period, inputting the depth representation of each view and the local adjacent graph into a second layer graph rolling network, extracting a clustering indication matrix of each view, training and optimizing the replacement matrix based on a unified loss function, and learning self-adaptive weights of the views to obtain a clustering result.","assignee":"National University of Defense Technology","inventors":["陶红","霍雨欣","江鸿宇"],"publication_date":"2026-02-13","filing_date":"2026-01-19","priority_date":"2026-01-19","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/70","G06V10/762","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/75","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/761","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121527463A/en"},{"publication_number":"CN121527097A","title":"A No-Reference Cloud Gaming Video Quality Assessment Method and Device Based on Local-Global Information Fusion","abstract":"本发明提供了一种基于局部全局信息融合的无参考云游戏视频质量评估方法及设备，包括以下步骤：步骤一、获取需要评估的云游戏视频，从整个视频帧中等间隔抽取K帧图像作为关键帧，步骤二、将抽取的关键帧分别送入深度特征提取支路和传统特征提取支路，对于深度特征提取支路，从关键帧中裁取指定大小的区域经过处理得到全局纹理特征，对于传统特征提取支路，由关键帧特定的时空域局部区域通过残差卷积网络获取局部深度特征，步骤三、将两支路得到的特征进行拼接并送入多层感知机质量回归器，得到最终的视频质量评分。本发明方案能够联合利用全局与局部信息，有效反映块效应、模糊及纹理失真等质量问题，实现高精度、轻量化的云游戏视频质量评估。 This invention provides a no-reference cloud gaming video quality assessment method and device based on local-global information fusion, comprising the following steps: Step 1, acquiring the cloud gaming video to be evaluated, and extracting K frames at equal intervals from the entire video frame as keyframes; Step 2, feeding the extracted keyframes into a depth feature extraction branch and a traditional feature extraction branch respectively. For the depth feature extraction branch, a specified size region is cropped from the keyframes and processed to obtain global texture features. For the traditional feature extraction branch, local depth features are obtained from specific spatiotemporal regions of the keyframes through a residual convolutional network; Step 3, concatenating the features obtained from the two branches and feeding them into a multilayer perceptron quality regressor to obtain the final video quality score. This invention can jointly utilize global and local information to effectively reflect quality problems such as blockiness, blurring, and texture distortion, achieving high-precision and lightweight cloud gaming video quality assessment.","assignee":"University of Electronic Science and Technology of China","inventors":["陆欣怡"],"publication_date":"2026-02-13","filing_date":"2026-01-19","priority_date":"2026-01-19","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/0002","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06T","G06T7/00","G06T7/40","G06T7/41","G","G06","G06V","G06V10/00","G06V10/20","G06V10/25","G","G06","G06V","G06V10/00","G06V10/40","G06V10/54","G","G06","G06V","G06V10/00","G06V10/70","G06V10/766","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10016","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20092","G06T2207/20104"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121527097A/en"},{"publication_number":"CN121525816A","title":"Knowledge graph construction method based on semantic network","abstract":"本发明涉及知识图谱构建技术领域，特别涉及一种基于语义网络的知识图谱构建方法。获取待构建知识图谱的生产数据集，采用安全知识识别器，进行安全知识识别，获得安全知识词向量集合；构建基础安全知识图谱；获得规模系数集合和获得占比系数集合；根据安全知识识别器的训练特征，分析获得事故系数集合、原因系数集合，对占比系数集合进行修正，处理获得事故标注系数集合和原因标注系数集合，对每个事故类型词向量和原因类型词向量的边尺度进行标注处理，获得安全知识图谱。本发明增加了安全知识图谱的可视化程度，实现了清晰呈现“事故类型－原因类型”的逻辑关联，并量化了事故严重程度、原因影响权重及两者的关联强度。 This invention relates to the field of knowledge graph construction technology, and particularly to a method for constructing a knowledge graph based on semantic networks. The method involves acquiring a production dataset for the knowledge graph to be constructed, using a security knowledge recognizer to identify security knowledge and obtain a set of security knowledge word vectors; constructing a basic security knowledge graph; obtaining a set of scale coefficients and a set of proportion coefficients; analyzing the training features of the security knowledge recognizer to obtain a set of accident coefficients and a set of cause coefficients; correcting the set of proportion coefficients; processing to obtain a set of accident annotation coefficients and a set of cause annotation coefficients; and annotating the side scales of each accident type word vector and cause type word vector to obtain the security knowledge graph. This invention increases the visualization of the security knowledge graph, clearly presenting the logical relationship between \"accident type - cause type,\" and quantifying the severity of the accident, the weight of the cause's influence, and the strength of the relationship between the two.","assignee":"Zhongan Huabang Beijing Safety Production Technology Research Institute Co ltd","inventors":["李进","赵守超","李军杰","樊延欣","杜锦涛"],"publication_date":"2026-02-13","filing_date":"2026-01-19","priority_date":"2026-01-19","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/02","G06N5/022"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121525816A/en"},{"publication_number":"CN121529851A","title":"Improved energy dispatch optimization method and system for grid-connected microgrid systems","abstract":"The invention provides an improved energy dispatching optimization method and system for a grid-connected micro-grid system, and relates to the technical field of micro-grid energy dispatching. According to the invention, the chaotic mapping Tent Map is firstly introduced into the algorithm to optimize the initialization of the algorithm, so that the lack of early diversity of MOBWO algorithm is avoided, the elite differential variation is secondly improved to be dynamic elite differential variation to replace the traditional variation operation in the BWO algorithm, the defect of the traditional mutation mechanism is effectively overcome, finally, the problem of insufficient local optimization of the algorithm is solved, a dynamic reverse learning mechanism is fused, reverse particles of the dynamic reverse learning mechanism can not only be used for carrying out reverse learning, but also the strong global searching capability of the BWO algorithm is utilized, and the solution quality is further improved.","assignee":"Anhui Jianzhu University","inventors":["黄梅初","展旭","窦艳","刘治国","谢飞翔"],"publication_date":"2026-02-13","filing_date":"2026-01-19","priority_date":"2026-01-19","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N7/00","G06N7/08","H","H02","H02J","H02J3/00","H02J3/17","H02J3/175","H","H02","H02J","H02J3/00","H02J3/28","H","H02","H02J","H02J3/00","H02J3/28","H02J3/32","H","H02","H02J","H02J3/00","H02J3/38","H02J3/46","H02J3/466","H","H02","H02J","H02J2101/00","H02J2101/10","H","H02","H02J","H02J2101/00","H02J2101/20","H","H02","H02J","H02J2101/00","H02J2101/20","H02J2101/22","H02J2101/24","H","H02","H02J","H02J2101/00","H02J2101/20","H02J2101/28","H","H02","H02J","H02J2103/00","H02J2103/30"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121529851A/en"},{"publication_number":"CN121524372A","title":"A hybrid approach of knowledge graphs and digital objects for scientific discovery","abstract":"本申请公开了一种用于科学发现的知识图谱与数字对象混合方法，属于数据检索领域，包括：对文献数据集进行解析，以获得文献数据集初始的概念数据集；对比初始的概念数据集中，每个概念数据的概念解释数据，以将表征为具有相同概念内涵的概念名称所对应的概念数据进行融合，以获得消除歧义后的概念数据集；根据消除歧义后的概念数据集，生成文献数据集的知识图谱数据。本申请通过联合存储‑概念消歧的核心思路，实现了从原始文献到结构化知识的逐层转化，进而使得语义检索误匹配和知识图谱缺乏文献精确对应的问题得到有效改善，从而提升了检索的准确性和效率性。 This application discloses a method for integrating knowledge graphs and digital objects in scientific discovery, belonging to the field of data retrieval. The method includes: parsing a document dataset to obtain an initial concept dataset; comparing the concept explanation data of each concept data in the initial concept dataset to fuse the concept data corresponding to concept names representing the same conceptual connotation, thereby obtaining an unambiguous concept dataset; and generating knowledge graph data of the document dataset based on the unambiguous concept dataset. This application, through the core idea of joint storage and concept disambiguation, achieves a layer-by-layer transformation from raw documents to structured knowledge, effectively improving the problems of semantic retrieval mismatches and the lack of precise document correspondence in knowledge graphs, thus enhancing the accuracy and efficiency of retrieval.","assignee":"Peking University","inventors":["景翔","黄东亮","孙保庆","黄罡"],"publication_date":"2026-02-13","filing_date":"2026-01-19","priority_date":"2026-01-19","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/38","G","G06","G06F","G06F16/00","G06F16/30","G06F16/31","G06F16/316","G","G06","G06F","G06F16/00","G06F16/30","G06F16/36","G06F16/367","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121524372A/en"},{"publication_number":"CN121525099A","title":"A method and system for privacy data protection in big data analysis process","abstract":"The invention discloses a method and a system for protecting private data in a big data analysis process, and relates to the technical field of civil aviation data safety and privacy protection. Extracting the association relation between fields through a convolutional neural network, calculating a field sensitivity initial value, generating a field sensitivity weight by combining a field uniqueness index and the sensitivity initial value, generating a privacy class mapping table containing the field sensitivity weight according to the sensitivity weight, calculating query sensitivity and privacy budget parameters, generating differential privacy noise parameters and homomorphic encryption keys, generating a hierarchical protection strategy, performing noise injection and ciphertext conversion on original data to generate a desensitized data set, performing passenger flow statistics through secret state calculation, and generating a publicly-available analysis result. The over protection of the low-sensitivity data is avoided, and the sufficient safety of the high-sensitivity data is ensured.","assignee":"Beijing Huacheng Zhiyun Software Co ltd","inventors":["娄健","张磊宏"],"publication_date":"2026-02-13","filing_date":"2026-01-19","priority_date":"2026-01-19","cpc_codes":["G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6245","G","G06","G06F","G06F21/00","G06F21/60","G06F21/602","G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6227","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121525099A/en"},{"publication_number":"AU2026200465A1","title":"Method and system for battery abnormality detection through artificial neural network battery model based on field data","abstract":"28 According to the present invention, there is provided a battery malfunctioning behavior detection method and system using field data for time-sequentially extracting real-time field data from a battery in operation, and detecting whether or not it is malfunctioning behavior of a battery according to the deviation by comparing prediction values of cell voltage and cell temperature from extracted field data with cell voltage and cell temperature values of real-time field data. 28 20 26 20 04 65 22 J an 2 02 6 J a n 2 0 2 6 A B S T R A C T 2 0 2 6 2 0 0 4 6 5 2 2","assignee":"LG Energy Solution Ltd","inventors":["Ki Wook Kwon","Jeong Bin Lee","Sung Yul YOON"],"publication_date":"2026-02-12","filing_date":"2026-01-22","priority_date":"2022-03-30","cpc_codes":["G","G01","G01R","G01R19/00","G01R19/0038","G","G01","G01R","G01R19/00","G01R19/165","G01R19/16533","G01R19/16538","G01R19/16542","G","G01","G01R","G01R19/00","G01R19/165","G01R19/16566","G01R19/16576","G","G01","G01R","G01R31/00","G01R31/36","G01R31/367","G","G01","G01R","G01R31/00","G01R31/36","G01R31/371","G","G01","G01R","G01R31/00","G01R31/36","G01R31/382","G","G01","G01R","G01R31/00","G01R31/36","G01R31/382","G01R31/3842","G","G01","G01R","G01R31/00","G01R31/36","G01R31/392","G","G01","G01R","G01R31/00","G01R31/36","G01R31/396","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","Y","Y02","Y02E","Y02E60/00","Y02E60/10"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026200465A1/en"},{"publication_number":"KR20260020455A","title":"Apparatus for forecasting of electric power based on a prediction model","abstract":"예측 모델 기반 전력 예측 장치가 개시된다. 본 발명의 일 측면에 따른 전예측 모델 기반 전력 예측 장치는, 과거 전력 시계열 데이터에 대해 전처리를 수행하여 학습용 데이터를 생성하는 전처리부, 제1 최적 하이퍼 파라미터를 탐색하고, 탐색된 제1 최적 하이퍼 파리미터에 기초하여 순환 신경망을 구성하며, 학습용 데이터를 순환 신경망에 입력하여 순환 신경망 모델을 학습하고, 학습된 순환 신경망 모델에 예측용 데이터를 입력하여 제1 시계열 데이터를 예측하는 순환 신경망 예측부, 제2 최적 하이퍼 파라미터를 탐색하고, 탐색된 제2 최적 하이퍼 파리미터에 기초하여 합성곱 신경망을 구성하며, 학습용 데이터를 합성곱 신경망에 입력하여 합성곱 신경망 모델을 학습하고, 학습된 합성곱 신경망 모델에 예측용 데이터를 입력하여 제2 시계열 데이터를 예측하는 합성곱 신경망 예측부, 및 학습용 데이터의 입력에 의한 순환 신경망 모델의 제1 출력값, 학습용 데이터의 입력에 의한 합성곱 신경망 모델의 제2 출력값 및 시계열 범주형 변수를 입력받아 완전연결 신경망 모델을 학습하고, 시계열 범주형 변수가 결합된 제1 시계열 데이터 및 제2 시계열 데이터를 완전연결 신경망 모델에 입력하여 최종 시계열 데이터를 예측하는 결합 예측부를 포함한다. A power prediction device based on a predictive model is disclosed. According to one aspect of the present invention, a power prediction device based on a pre-prediction model comprises: a preprocessing unit that performs preprocessing on past power time series data to generate training data; a recurrent neural network prediction unit that searches for a first optimal hyper parameter, constructs a recurrent neural network based on the searched first optimal hyper parameter, inputs training data into the recurrent neural network to train the recurrent neural network model, and inputs prediction data into the trained recurrent neural network model to predict the first time series data; a convolutional neural network prediction unit that searches for a second optimal hyper parameter, constructs a convolutional neural network based on the searched second optimal hyper parameter, inputs training data into the convolutional neural network to train the convolutional neural network model, and inputs prediction data into the trained convolutional neural network model to predict the second time series data; and a first output value of the recurrent neural network model by inputting training data, a second output value of the convolutional neural network model by inputting training data, and a time series categorical variable are input to train a fully connected neural network model, and inputs the first time series data and the second time series data in which time series categorical variables are combined into the fully connected neural network model. Includes a combined prediction unit that inputs and predicts the final time series data.","assignee":"한국전력공사","inventors":["이정일","최윤아","정남준","장민영"],"publication_date":"2026-02-11","filing_date":"2026-02-02","priority_date":"2022-08-08","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06Q","G06Q50/00","G06Q50/06","H","H02","H02J","H02J3/00","H02J3/003","H","H02","H02J","H02J3/00","H02J3/004"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260020455A/en"},{"publication_number":"KR20260020161A","title":"Method to predict heart age","abstract":"본 개시의 일 실시예에 따라 사전 학습된 인공 신경망 모델을 활용하여 심장 나이를 추정하는 방법이 개시된다. 구체적으로, 본 개시에 따르면, 컴퓨팅 장치가, 사용자의 생체 신호 데이터를 획득하고, 사전 학습된 인공 신경망 모델을 활용하여, 상기 사용자의 생체 신호 데이터를 기초로 상기 사용자의 심장 나이를 추정하고, 상기 추정된 사용자의 심장 나이와 연관된 분석 정보 또는 미래 예측 정보 중 적어도 하나의 정보를 생성할 수 있다. According to one embodiment of the present disclosure, a method for estimating heart age using a pre-trained artificial neural network model is disclosed. Specifically, according to the present disclosure, a computing device may acquire a user's biosignal data, estimate the user's heart age based on the user's biosignal data using a pre-trained artificial neural network model, and generate at least one piece of information, either analysis information or future prediction information, associated with the estimated user's heart age.","assignee":"주식회사 뷰노","inventors":["주성훈","장민옥","김경근","이성재","나영연"],"publication_date":"2026-02-10","filing_date":"2026-01-22","priority_date":"2022-08-18","cpc_codes":["A","A61","A61B","A61B5/00","A61B5/02","G","G16","G16H","G16H50/00","G16H50/50","A","A61","A61B","A61B5/00","A61B5/02","A61B5/02007","A","A61","A61B","A61B5/00","A61B5/02","A61B5/0205","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7275","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H15/00","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/70","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260020161A/en"},{"publication_number":"KR20260019578A","title":"A real-time investment speed automatic decision system based on generative AI for stock and coin investment, and an investment management platform utilizing this system","abstract":"본 발명은 주식 및 가상자산(코인) 투자에서 시장 데이터 및 투자자 거래 패턴을 기반으로 밀리초(ms) 단위의 주문 실행속도(Execution Speed)에 해당하는 투자속도 S(t)를 자동 산출하고, 산출된 S(t)에 따라 주문 빈도, 주문 규모, 진입·청산 시점 및 포지션 비중을 자동 결정하여 주문 실행 및 투자관리를 실시간으로 수행하는 AI 기반 실시간 투자속도 자동결정 시스템 및 이를 이용한 투자관리 플랫폼에 관한 것이다. 본 발명에 따르면 급변 시장에서 반응성을 향상시키고 과도한 주문 실행을 억제하며 슬리피지를 완화하여 투자 리스크를 경감하고 실행 효율을 향상시키는 효과가 있다. The present invention relates to an AI-based real-time investment speed automatic determination system and an investment management platform using the same, which automatically calculates an investment speed S(t), which corresponds to an order execution speed in milliseconds (ms) based on market data and investor trading patterns in stock and virtual asset (coin) investments, and automatically determines order frequency, order size, entry/liquidation points, and position ratio based on the calculated S(t), thereby executing orders and managing investments in real time. According to the present invention, there is an effect of improving responsiveness in a rapidly changing market, suppressing excessive order execution, and alleviating slippage, thereby reducing investment risk and improving execution efficiency.","assignee":"심재훈","inventors":["심재훈"],"publication_date":"2026-02-10","filing_date":"2026-01-15","priority_date":"2024-04-23","cpc_codes":["G","G06","G06Q","G06Q40/00","G06Q40/06","G06Q40/063","G06Q40/0631","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06Q","G06Q10/00","G06Q10/10","G","G06","G06Q","G06Q40/00","G06Q40/04","G","G06","G06Q","G06Q40/00","G06Q40/04","G06Q40/042","G06Q40/0421","G","G06","G06Q","G06Q40/00","G06Q40/04","G06Q40/043","G","G06","G06Q","G06Q40/00","G06Q40/04","G06Q40/045","G","G06","G06Q","G06Q40/00","G06Q40/06","H","H04","H04L","H04L9/00","H04L9/50"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260019578A/en"},{"publication_number":"CN121500226A","title":"A fault detection method and system based on smart meters","abstract":"The application provides a fault detection method and a fault detection system based on an intelligent ammeter, which relate to the technical field of intelligent ammeter fault detection and diagnosis and have the technical scheme that three-dimensional data are obtained, wherein the three-dimensional data comprise intelligent ammeter data, environment sensing data, power grid topology data and line load data; and predicting a normal data range of 1 minute in the future in real time according to the three-dimensional data, judging whether actual data exceeds the normal data range, uploading all data to a cloud end for analyzing and detecting faults when the actual data exceeds the normal data range, and collecting fault data in real time and generating a fault early warning list. The fault detection method and system based on the intelligent ammeter have the advantages of reducing fault false alarm rate, strengthening fault detection response efficiency and realizing fault risk early warning.","assignee":"Wuhan Youxunda Technology Co ltd; Shenzhen Friendcom Technology Co Ltd","inventors":["庞景夏","乐渭斌","张博涛","韩安孟","李颂清"],"publication_date":"2026-02-10","filing_date":"2026-01-14","priority_date":"2026-01-14","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G01","G01R","G01R35/00","G01R35/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121500226A/en"},{"publication_number":"CN121503522A","title":"Conveyor Belt Bottle Cap Counting System Based on Deep Learning and Multi-Object Tracking","abstract":"本发明涉及工业自动化与机器视觉技术领域，公开了基于深度学习与多目标跟踪的传送带瓶盖计数系统，所述系统包括：图像采集模块，用于通过轻量封装的海康GigE相机SDK实现设备枚举、包长优化、关闭触发、持续取流以及多种像素格式的统一转换，输出稳定的BGR图像帧；检测与跟踪模块，用于加载预训练的YOLO模型对图像帧进行瓶盖目标检测，本发明利用YOLO模型对瓶盖目标进行检测，并通过多目标跟踪算法为每个检测目标分配唯一的track id，在此基础上引入基于水平基准线与目标中心点历史的跨线计数策略，仅当目标中心点在相邻图像帧之间跨越虚拟计数线时才触发计数事件。 This invention relates to the fields of industrial automation and machine vision technology, and discloses a conveyor belt bottle cap counting system based on deep learning and multi-object tracking. The system includes: an image acquisition module, used to implement device enumeration, packet length optimization, shutdown triggering, continuous streaming, and unified conversion of multiple pixel formats through a lightweight Hikvision GigE camera SDK, and output stable BGR image frames; and a detection and tracking module, used to load a pre-trained YOLO model to detect bottle cap targets on the image frames. This invention uses the YOLO model to detect bottle cap targets and assigns a unique track ID to each detected target through a multi-object tracking algorithm. On this basis, a cross-line counting strategy based on the horizontal baseline and the history of the target center point is introduced, and a counting event is triggered only when the target center point crosses a virtual counting line between adjacent image frames.","assignee":"Changchun University","inventors":["郭锐强","丛达","张贺","贾力苏","林笑宇","乔秀芳","张洋","章亚频","全家乐","孙文奇","刘文龙","李雪见","纪培勇","胡景琦"],"publication_date":"2026-02-10","filing_date":"2026-01-14","priority_date":"2026-01-14","cpc_codes":["G","G06","G06M","G06M7/00","G06M7/02","G06M7/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06V","G06V10/00","G06V10/10","G","G06","G06V","G06V10/00","G06V10/40","G06V10/62","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V2201/00","G06V2201/07","Y","Y02","Y02P","Y02P90/00","Y02P90/30"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121503522A/en"},{"publication_number":"CN121503827A","title":"A Self-Evolving Control System and Method for Mines Based on Causal Discovery and Meta-Reinforcement Learning","abstract":"本发明涉及矿山智能调控技术领域，公开了基于因果发现与元强化学习的矿山自演进调控系统及方法，包括数据采集与预处理模块、因果发现与嵌套图构建模块、策略元学习与协进化扩散模块、自主演化执行与调控模块以及系统迭代更新模块；其中，所述数据采集与预处理模块从矿山环境中采集多源数据，并应用模态共振融合机制生成融合数据集。本发明中，通过数据采集与预处理模块应用模态共振融合机制生成融合数据集，增强多源数据间的隐含关系提取，结合因果发现与嵌套图构建模块生成初始自相似因果嵌套图，精确捕获不同尺度的因果链，为后续策略生成提供高保真输入。 This invention relates to the field of intelligent mine control technology, and discloses a mine self-evolutionary control system and method based on causal discovery and meta-reinforcement learning. The system includes a data acquisition and preprocessing module, a causal discovery and nested graph construction module, a policy meta-learning and co-evolutionary diffusion module, an autonomous evolutionary execution and control module, and a system iterative update module. The data acquisition and preprocessing module collects multi-source data from the mine environment and applies a modal resonance fusion mechanism to generate a fused dataset. In this invention, the data acquisition and preprocessing module uses a modal resonance fusion mechanism to generate a fused dataset, enhancing the extraction of implicit relationships between multi-source data. Combined with the causal discovery and nested graph construction module, an initial self-similar causal nested graph is generated, accurately capturing causal chains at different scales and providing high-fidelity input for subsequent policy generation.","assignee":"Geological Disaster Prevention And Control Center Of Henan Provincial Geological Bureau; Jinggui Intelligence Henan Intelligent Equipment Co ltd","inventors":["贾尚伟","贺富领","王军","衡家然","王小龙","娄涛","王海洋","杨亚威","崔锦康"],"publication_date":"2026-02-10","filing_date":"2026-01-14","priority_date":"2026-01-14","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G","G06","G06Q","G06Q50/00","G06Q50/02","Y","Y02","Y02P","Y02P90/00","Y02P90/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121503827A/en"},{"publication_number":"CN121508695A","title":"A method and system for evaluating the synergistic effect of multi-target electromagnetic interference","abstract":"The invention relates to the technical field of electromagnetic environmental effect evaluation and electronic countermeasure, in particular to a multi-target electromagnetic interference synergistic effect evaluation method and system, comprising the steps of obtaining waveform data of a wide area coverage vector and a time domain pulse vector; the method comprises the steps of obtaining total radio frequency energy, calculating a vulnerable window of a protocol state machine, calculating optimal cooperative delay, collecting link state data of tested equipment, calculating protocol state entropy based on the link state data, calculating a cooperative efficiency index by combining the variation of the protocol state entropy and the total radio frequency energy, correcting the optimal cooperative delay by using a gradient descent algorithm and returning to execute a transmitting step if the variation of the cooperative efficiency index is larger than a preset convergence threshold, and solidifying a wide area coverage vector, a time domain pulse vector and the optimal cooperative delay into a standard test case if the variation of the cooperative efficiency index is smaller than or equal to the preset convergence threshold.","assignee":"Shenghang Taizhou Technology Co ltd","inventors":["常选明","张镭","顾晓乐"],"publication_date":"2026-02-10","filing_date":"2026-01-14","priority_date":"2026-01-14","cpc_codes":["H","H04","H04B","H04B17/00","H04B17/30","H04B17/309","H04B17/345","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06N","G06N20/00","G","G06","G06F","G06F2123/00","G06F2123/02","G","G06","G06F","G06F2218/00","G06F2218/08","Y","Y02","Y02D","Y02D30/00","Y02D30/70"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121508695A/en"},{"publication_number":"CN121503306A","title":"A Simulation Method and System for Flood Resilience Evolution Based on the Interaction of Natural and Social Elements","abstract":"The invention discloses a flood control toughness evolution simulation method and system based on natural-social element interaction, wherein the method comprises the steps of obtaining a natural element space-time sequence and a social element space-time sequence of a target area, and aligning the natural element space-time sequence and the social element space-time sequence to uniform space-time grid nodes to obtain a heterogeneous node sequence; the method comprises the steps of calculating time-varying interaction intensity of nodes along with time variation and interaction uncertainty of the time-varying interaction intensity, compiling and generating time-varying rule parameters between each pair of interaction nodes by using a preset rule compiler, inputting the time-varying rule parameters into a pre-constructed toughness dynamics model, and driving and updating toughness state values of each node in continuous time steps to obtain a flood control toughness evolution track. The method solves the problems of data driving and mechanism model splitting in the prior art through a rule compiling mechanism, realizes the dynamic conversion of interaction characteristics from soft weights to hard rules, and effectively improves the accuracy and the interpretability of flood control toughness assessment under complex time-varying situations.","assignee":"Nanjing Hydraulic Research Institute of National Energy Administration Ministry of Transport Ministry of Water Resources","inventors":["苏鑫","王莉莉","王磊之","李伶杰","刘勇","胡鉴闻","张野","陈兆懿","轩省伟","云兆得","王双","程慧宇","方华荣","周庆霈","徐致扬"],"publication_date":"2026-02-10","filing_date":"2026-01-14","priority_date":"2026-01-14","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06F","G06F2119/00","G06F2119/14","Y","Y02","Y02A","Y02A10/00","Y02A10/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121503306A/en"},{"publication_number":"KR20260018956A","title":"Service system based on distributed calculation","abstract":"본 발명은 분산 처리 기반 서비스 시스템에 관한 것으로, 인공지능 학습 단계에서 변수와 파라미터를 최소화하여 조건에 맞는 경량화 함수를 자동으로 생성하고, 분산병렬 연산 장치를 이용하여 인공지능 분석모델을 연산처리하고, 복수의 경량화된 인공지능 서비스를 동시에 사용하여 경량화 인공지능 서비스를 제공함으로써, 능동적인 인공지능 플랫폼 역할을 수행하며, 인공지능 분석모델을 손쉽게 경량화하고 빠르고 정확도 높은 서비스를 제공할 수 있고, 개발된 인공지능 어플리케이션의 정확도를 측정하여 품질을 관리하고 체계적으로 성능을 관리하며, 표준화된 인터페이스를 통해 다수의 어플리케이션에서 동일한 인공지능 서비스를 사용할 수 있도록 한다. The present invention relates to a distributed processing-based service system, which automatically generates a lightweight function that satisfies conditions by minimizing variables and parameters in an artificial intelligence learning stage, computes an artificial intelligence analysis model using a distributed parallel computing device, and provides a lightweight artificial intelligence service by simultaneously using multiple lightweight artificial intelligence services, thereby serving as an active artificial intelligence platform, and can easily lightweight an artificial intelligence analysis model and provide a fast and highly accurate service, and manages quality and systematically manages performance by measuring the accuracy of a developed artificial intelligence application, and enables the use of the same artificial intelligence service in multiple applications through a standardized interface.","assignee":"한국전력공사","inventors":["장민영","정남준","성창환","이정일"],"publication_date":"2026-02-09","filing_date":"2026-01-29","priority_date":"2022-08-17","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260018956A/en"},{"publication_number":"NL4000335A","title":"A blind-assistance target detection method based on a cascaded feature pyramid and lightweight design","abstract":"The invention belongs to the field of computer vision and provides a blindassistance target detection method based on a cascaded feature pyramid and lightweight design, comprising the following steps: acquiring the PASCAL VOC2007 public dataset; reconstructing the neck part of YOLOl 1 using an SBA module, which enhances the model’ s performance in small-object detection and multi-scale target detection tasks; replacing the Bottleneck in C3k2 of the original model with Bottleneck_DEConv, and substituting the second Conv layer with a C3K2-DEConv module composed of multiple parallel C3k_DECOnv submodules. Through residual connections, the transmission capability of detailed information is strengthened. Finally, adopting the Adown module, which is mainly inspired by the lightweight “average pooling + max pooling parallel” dual-path downsampling strategy, achieves reduced computation through downsampling and parameter reduction, replacing the original downsampling module to lower the model’ 3 complexity, thereby obtaining the KJ -YOLO model. This invention effectively improves the detection capability of blind-assistance devices in complex environments and addresses the problem of missed detection of small targets, enhancing the detection accuracy of such devices and offering better practicality and adaptability.","assignee":"Guangdong Univ Science And Technology","inventors":["Chen Yuyi","Liu Yuzhou","Gong Shu","Qiu Caihua","Hao Yilu"],"publication_date":"2026-02-09","filing_date":"2025-11-10","priority_date":"2025-10-21","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06V","G06V10/00","G06V10/40","G06V10/46","G06V10/469","G06V10/473","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/776","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V2201/00","G06V2201/07"],"country":"NL","kind":"application","source_url":"https://patents.google.com/patent/NL4000335A/en"},{"publication_number":"LU602828B1","title":"Group logistics transportation scheduling method and system based on role interaction graph neural network","abstract":"The present invention relates to the field of combinatorial optimization and artificial intelligence, and discloses a group logistics transportation scheduling method and system based on a role-interaction graph neural network. The method comprises the following steps: S1, dividing agent nodes and location nodes and generating initial features; S2, iteratively updating node embeddings using a multi-channel attention mechanism in a graph neural network; S3, generating delivery point allocation probabilities based on node embeddings and determining the initial allocation scheme; S4, performing local reallocation optimization for delivery points with low confidence allocations; S5, planning paths in parallel for the optimal scheme, outputting the results and using them for reinforcement learning feedback. In the present invention, by modeling complex interaction relationships through a multi-channel attention mechanism in a graph neural network, and through the synergy of local reallocation optimization and parallel path planning, high-quality scheduling schemes can be generated within seconds.","assignee":"Changan Univ","inventors":["Yujiao Hu"],"publication_date":"2026-02-09","filing_date":"2025-08-07","priority_date":"2025-07-01","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06311","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/083","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/083","G06Q10/08355","Y","Y02","Y02T","Y02T10/00","Y02T10/10","Y02T10/40"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU602828B1/en"},{"publication_number":"LU602820B1","title":"Anchor bolt hole positioning method and system based on neural network and grey wolf optimizer","abstract":"This disclosure relates to the technical field of machine vision and image processing, and specifically provides an anchor bolt hole positioning method and system based on a neural network and a grey wolf optimizer. The method includes: obtaining a two- dimensional image of an inner wall of a tunnel; performing uniform segmentation on the two-dimensional image of the inner wall of the tunnel to obtain a plurality of square areas and establish a coordinate system; determining whether a preset anchor bolt hole of a tunnel catenary avoids a crack, a water seepage point, and other unfavorable conditions, if so, determining a current anchor bolt hole position, and if not, randomly generating artificial wolves in the square area with the coordinate system by using the grey wolf optimizer; establishing a fitness function for performing wolf sorting to obtain an alpha wolf, a scout wolf, and a fierce wolf.","assignee":"China Constr Ind & Energy Eng Group Co Ltd; Univ Tianjin Chengjian; China Construction Rail Electrification Eng Co Ltd","inventors":["Wei Li","Jianbing Gu","Yi Pang","Mingming Zhang","Liyang Zhang","Rui Gao","Jing Liu","Kai Cheng","Lei Pan","Jianwei Chen","Guohua Zhang","Qi Li"],"publication_date":"2026-02-09","filing_date":"2025-08-07","priority_date":"2025-03-26","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/70","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU602820B1/en"},{"publication_number":"KR20260018739A","title":"A method and an electronic device for performing convolutional neural network operations on a batch of input images using homomorphic encryption","abstract":"동형암호를 이용하여 복수의 입력 이미지 배치에 대한 컨볼루션 신경망(convolutional neural network) 연산을 수행하는 방법 및 전자 장치가 제공된다. 본 방법은 입력 이미지 배치의 각 채널에 대하여, 배치에 포함된 이미지들에서 동일한 위치의 픽셀 값들을 하나의 벡터로 그룹화하고, 그룹화된 벡터들을 동형암호 방식으로 암호화하여 제1 암호문을 획득하는 단계, 각 채널 및 암호문의 각 계수에 대하여, 모든 픽셀 위치에 대응하는 제1 암호문의 계수들을 그룹화하여 다항식을 생성하고, 다항식들을 하나의 행렬로 구성하는 단계, 행렬을 컨볼루션 필터를 나타내는 다항식 행렬과 곱하여 결과 행렬을 획득하고, 결과 행렬의 각 요소에 대해 모듈로(modulo) 연산을 수행하여 암호화된 컨볼루션 연산 결과를 생성하는 단계. 여기서, 연산은 암호화되지 않는 행렬들에 대해 수행되며, 결과 행렬의 다항식 계수를 재배열하여, 컨볼루션 레이어가 적용된 후의 이미지 배치에 대한 제2 암호문을 생성하는 단계를 포함하며, 제2 암호문은 제1 암호문과 동일한 동형 암호화 방식으로 암호화된다. A method and an electronic device for performing a convolutional neural network operation on a batch of multiple input images using homomorphic encryption are provided. The method comprises the steps of: for each channel of an input image batch, grouping pixel values at the same location in images included in the batch into a single vector; and encrypting the grouped vectors using a homomorphic encryption method to obtain a first ciphertext; for each channel and each coefficient of the ciphertext, grouping coefficients of the first ciphertext corresponding to all pixel locations to generate a polynomial; and configuring the polynomials into a single matrix; multiplying the matrix by a polynomial matrix representing a convolution filter to obtain a result matrix; and performing a modulo operation on each element of the result matrix to generate an encrypted convolution operation result. Here, the operation is performed on unencrypted matrices, and the step of rearranging the polynomial coefficients of the result matrix to generate a second ciphertext for the image batch after the convolution layer is applied, wherein the second ciphertext is encrypted using the same homomorphic encryption method as the first ciphertext.","assignee":"주식회사 크립토랩","inventors":["기욤 앙로","시메옹 라포르트"],"publication_date":"2026-02-09","filing_date":"2025-07-30","priority_date":"2024-07-31","cpc_codes":["H","H04","H04L","H04L9/00","H04L9/008","G","G06","G06F","G06F17/00","G06F17/10","G06F17/14","G06F17/141","G","G06","G06F","G06F17/00","G06F17/10","G06F17/16","G","G06","G06F","G06F7/00","G06F7/60","G06F7/72","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260018739A/en"},{"publication_number":"KR20260018710A","title":"Artificial intelligence application development system","abstract":"본 발명은 그래픽 기반으로 인공지능 어플리케이션을 용이하게 개발할 수 있도록 해주는 인공지능 어플리케이셔 개발 시스템에 관한 것이다. 본 발명에 따른 인공지능 어플리케이션 개발 시스템은 개발자가 그래픽 베이스로 어플리케이션을 개발할 수 있는 IDE를 제공하는 IDE 제공장치(1)와, 노드 라이브러리와 프로젝트 라이브러리를 구비하는 저장장치를 포함하여 구성된다. IDE 제공장치(1)는 개발자가 노드와 엣지를 이용하여 흐름도를 생성할 수 있는 사용자 인터페이스(20)를 구비하고, 사용자 인터페이스(20)는 개발자가 노드를 작업공간으로 호출하기 위한 노드 호출수단과, 노드 사이에 프로세스 흐름 또는 데이터 흐름을 설정하기 위한 엣지를 연결하기 위한 엣지 연결수단을 구비한다. 또한, IDE 제공장치(1)는 개발자가 작업공간으로 호출한 노드의 노드 정보를 편집하기 위한 노드 편집수단을 제공한다. The present invention relates to an artificial intelligence application development system that enables easy development of artificial intelligence applications based on graphics. The artificial intelligence application development system according to the present invention comprises an IDE providing device (1) that provides an IDE for developers to develop applications based on graphics, and a storage device having a node library and a project library. The IDE providing device (1) has a user interface (20) that allows developers to create flowcharts using nodes and edges, and the user interface (20) has a node calling means for developers to call nodes into a workspace, and an edge connecting means for connecting edges to establish a process flow or data flow between nodes. In addition, the IDE providing device (1) provides a node editing means for developers to edit node information of nodes called into a workspace.","assignee":"주식회사 미리내테크놀로지스","inventors":["유환수","존 웨인라이트"],"publication_date":"2026-02-09","filing_date":"2025-07-23","priority_date":"2024-07-31","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/10","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481","G06F3/0482","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G06F3/0486","G","G06","G06F","G06F8/00","G06F8/30","G06F8/34","G","G06","G06F","G06F8/00","G06F8/30","G06F8/38"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260018710A/en"},{"publication_number":"KR20260018709A","title":"Artificial intelligence application development system","abstract":"본 발명은 그래픽 기반으로 인공지능 어플리케이션을 용이하게 개발할 수 있도록 해주는 인공지능 어플리케이셔 개발 시스템에 관한 것이다. 본 발명에 따른 인공지능 어플리케이션 개발 시스템은 개발자가 그래픽 베이스로 어플리케이션을 개발할 수 있는 IDE를 제공하는 IDE 제공장치(1)와, 노드 라이브러리와 프로젝트 라이브러리를 구비하는 저장장치를 포함하여 구성된다. IDE 제공장치(1)는 개발자가 노드와 엣지를 이용하여 흐름도를 생성할 수 있는 사용자 인터페이스(20)를 구비하고, 사용자 인터페이스(20)는 개발자가 노드를 작업공간으로 호출하기 위한 노드 호출수단과, 노드 사이에 프로세스 흐름 또는 데이터 흐름을 설정하기 위한 엣지를 연결하기 위한 엣지 연결수단을 구비한다. 또한, IDE 제공장치(1)는 노드에 대해 하이퍼파라미터를 설정하기 위한 하이퍼하라미터 엔진(13)과, 흐름도를 해석하고 실행하기 위한 흐름도 엔진(12)을 포함한다. The present invention relates to an artificial intelligence application development system that enables easy development of artificial intelligence applications based on graphics. The artificial intelligence application development system according to the present invention comprises an IDE providing device (1) that provides an IDE for developers to develop applications based on graphics, and a storage device having a node library and a project library. The IDE providing device (1) has a user interface (20) that allows developers to create flowcharts using nodes and edges, and the user interface (20) has a node calling means for developers to call nodes into a workspace, and an edge connecting means for connecting edges to establish process flows or data flows between nodes. In addition, the IDE providing device (1) includes a hyperparameter engine (13) for setting hyperparameters for nodes, and a flowchart engine (12) for interpreting and executing flowcharts.","assignee":"주식회사 미리내테크놀로지스","inventors":["유환수","존 웨인라이트"],"publication_date":"2026-02-09","filing_date":"2025-07-23","priority_date":"2024-07-31","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/10","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G06F3/04845","G","G06","G06F","G06F8/00","G06F8/30","G06F8/34","G","G06","G06F","G06F8/00","G06F8/30","G06F8/36","G","G06","G06F","G06F8/00","G06F8/30","G06F8/38","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260018709A/en"},{"publication_number":"KR20260018707A","title":"Artificial intelligence application development system","abstract":"본 발명은 그래픽 기반으로 인공지능 어플리케이션을 용이하게 개발할 수 있도록 해주는 인공지능 어플리케이셔 개발 시스템에 관한 것이다. 본 발명에 따른 인공지능 어플리케이션 개발 시스템은 개발자가 그래픽 베이스로 어플리케이션을 개발할 수 있는 IDE를 제공하는 IDE 제공장치(1)와, 노드 라이브러리와 프로젝트 라이브러리를 구비하는 저장장치를 포함하여 구성된다. IDE 제공장치(1)는 개발자가 노드와 엣지를 이용하여 흐름도를 생성할 수 있는 사용자 인터페이스(20)를 구비하고, 사용자 인터페이스(20)는 개발자가 노드를 작업공간으로 호출하기 위한 노드 호출수단과, 노드 사이에 프로세스 흐름 또는 데이터 흐름을 설정하기 위한 엣지를 연결하기 위한 엣지 연결수단을 구비한다. 또한, IDE 제공장치(1)는 노드에 대해 하이퍼파라미터를 설정하기 위한 하이퍼하라미터 엔진(13)과, 흐름도를 해석하고 실행하기 위한 흐름도 엔진(12)을 포함한다. The present invention relates to an artificial intelligence application development system that enables easy development of artificial intelligence applications based on graphics. The artificial intelligence application development system according to the present invention comprises an IDE providing device (1) that provides an IDE for developers to develop applications based on graphics, and a storage device having a node library and a project library. The IDE providing device (1) has a user interface (20) that allows developers to create flowcharts using nodes and edges, and the user interface (20) has a node calling means for developers to call nodes into a workspace, and an edge connecting means for connecting edges to establish process flows or data flows between nodes. In addition, the IDE providing device (1) includes a hyperparameter engine (13) for setting hyperparameters for nodes, and a flowchart engine (12) for interpreting and executing flowcharts.","assignee":"주식회사 미리내테크놀로지스","inventors":["유환수","존 웨인라이트"],"publication_date":"2026-02-09","filing_date":"2025-07-22","priority_date":"2024-07-31","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/10","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481","G06F3/0482","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G06F3/0486","G","G06","G06F","G06F8/00","G06F8/30","G06F8/34","G","G06","G06F","G06F8/00","G06F8/30","G06F8/38"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260018707A/en"},{"publication_number":"KR20260018957A","title":"Service method based on distributed calculation","abstract":"본 발명은 분산 처리 기반 서비스 방법에 관한 것으로, 인공지능 학습 단계에서 변수와 파라미터를 최소화하여 조건에 맞는 경량화 함수를 자동으로 생성하고, 분산병렬 연산 장치를 이용하여 인공지능 분석모델을 연산처리하고, 복수의 경량화된 인공지능 서비스를 동시에 사용하여 경량화 인공지능 서비스를 제공함으로써, 능동적인 인공지능 플랫폼 역할을 수행하며, 인공지능 분석모델을 손쉽게 경량화하고 빠르고 정확도 높은 서비스를 제공할 수 있고, 개발된 인공지능 어플리케이션의 정확도를 측정하여 품질을 관리하고 체계적으로 성능을 관리하며, 표준화된 인터페이스를 통해 다수의 어플리케이션에서 동일한 인공지능 서비스를 사용할 수 있도록 한다. The present invention relates to a distributed processing-based service method, which automatically generates a lightweight function that meets conditions by minimizing variables and parameters in an artificial intelligence learning stage, computes an artificial intelligence analysis model using a distributed parallel computing device, and provides a lightweight artificial intelligence service by simultaneously using multiple lightweight artificial intelligence services, thereby serving as an active artificial intelligence platform, and can easily lightweight an artificial intelligence analysis model and provide a fast and highly accurate service, and manages quality and systematically manages performance by measuring the accuracy of a developed artificial intelligence application, and enables the use of the same artificial intelligence service in multiple applications through a standardized interface.","assignee":"한국전력공사","inventors":["장민영","정남준","성창환","이정일"],"publication_date":"2026-02-09","filing_date":"2026-01-29","priority_date":"2022-08-17","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260018957A/en"},{"publication_number":"KR20260018704A","title":"Apparatus, method, and computer program for training neural network model based on training data obtained through self-labeling method","abstract":"본 개시는 셀프 라벨링을 통해 획득된 학습 데이터에 기초하여 신경망 모델을 학습시키는 장치, 방법 및 컴퓨터 프로그램을 제공한다. 본 개시의 일 실시 예에 따른 상기 방법은 셀프 라벨링(Self-Labeling) 방식에 따라 생체 데이터에 라벨을 부여하고, 상기 라벨이 부여된 생체 데이터에 기반한 학습 데이터를 획득하는 단계와 상기 획득된 학습 데이터에 기초하여 신경망 모델을 학습 시키는 단계를 포함한다. The present disclosure provides a device, method, and computer program for training a neural network model based on training data acquired through self-labeling. According to one embodiment of the present disclosure, the method comprises the steps of labeling biometric data using a self-labeling method, acquiring training data based on the labeled biometric data, and training a neural network model based on the acquired training data.","assignee":"주식회사 메디컬에이아이","inventors":["권준명","장종환"],"publication_date":"2026-02-09","filing_date":"2025-07-21","priority_date":"2024-07-31","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/70","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/318","A61B5/346","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H50/00","G16H50/50"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260018704A/en"},{"publication_number":"KR102924588B1","title":"Ai-based smart building exterior wall and window cleaning drone and cleaning system including the same","abstract":"본 발명은 구조물 청소 시스템에 관한 것으로, 보다 상세하게는, 구조물의 3D 모델링을 통해 BIM(Building Information Modeling)을 완성하여 오염을 식별하고, RTK(Real Time Kinematic) 센서를 장착한 드론을 통해 구조물의 외벽을 자율비행으로 청소할 수 있는 드론 및 이를 포함하는 청소 시스템에 관한 것이다. 상기 청소 시스템은, 드론, 상기 드론과 물리적으로 연결되어 상기 드론에 액체를 전달하는 호스, 상기 호스를 제어하는 제1 윈치 및 상기 호스에 액체를 전달하는 워터 탱크와 세정제 탱크를 포함하는 제공부, 상기 호스와 연결되는 로프 및 상기 로프를 제어하는 제2 윈치를 포함하는 지지부 및 상기 드론이 상기 로프와 구조적으로 연결되는 상기 호스로부터 제공되는 액체를 이용하여 상기 구조물에 대한 청소 작업을 수행하도록 제어하는 제어 서버를 포함할 수 있다. The present invention relates to a structure cleaning system, and more specifically, to a drone capable of cleaning the outer wall of a structure through autonomous flight using a drone equipped with an RTK (Real Time Kinematic) sensor, by completing BIM (Building Information Modeling) through 3D modeling of the structure to identify contamination, and a cleaning system including the same. The cleaning system may include a drone, a hose physically connected to the drone to deliver liquid to the drone, a first winch controlling the hose, and a provision unit including a water tank and a detergent tank for delivering liquid to the hose, a support unit including a rope connected to the hose and a second winch controlling the rope, and a control server controlling the drone to perform a cleaning operation on the structure using liquid provided from the hose structurally connected to the rope.","assignee":"(주)플라이존드론","inventors":["안승용"],"publication_date":"2026-02-09","filing_date":"2025-05-07","priority_date":"2025-05-07","cpc_codes":["B","B64","B64D","B64D1/00","B64D1/16","B64D1/18","B","B05","B05B","B05B12/00","B05B12/08","B05B12/12","B","B05","B05B","B05B13/00","B05B13/005","B","B64","B64U","B64U20/00","B64U20/80","B64U20/87","E","E04","E04G","E04G23/00","E04G23/002","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06T","G06T17/00","G06T17/20","B","B64","B64U","B64U2101/00","B64U2101/25","B64U2101/29","B","B64","B64U","B64U2101/00","B64U2101/30","B","B64","B64U","B64U2101/00","B64U2101/45","B","B64","B64U","B64U2201/00","B64U2201/10"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102924588B1/en"},{"publication_number":"KR20260018141A","title":"Method and system for perfomring tasks based on tabular data analysis","abstract":"일 실시예는, 컴퓨터가 수행하는 방법으로서, 분석용 테이블형 데이터를 획득하여 적어도 하나의 메모리에 저장하는 단계, 적어도 하나의 프로세서가 제1 인공지능 모델을 이용하여 상기 분석용 테이블형 데이터를 인코딩하여 분석용 잠재 변수 데이터를 생성하는 단계, 여기서 상기 제1 인공지능 모델은 학습용 테이블형 데이터를 기초로 하여 학습용 잠재 변수 데이터를 생성하고, 상기 학습용 잠재 변수 데이터를 기초로 하여 상기 학습용 테이블형 데이터에 대한 구간화 데이터를 복원하도록 사전 학습되고, 상기 적어도 하나의 프로세서가 상기 분석용 잠재 변수 데이터를 기초로 제2 인공지능 모델을 이용하여 상기 분석용 테이블형 데이터에 대한 분석 결과를 생성하는 단계, 및 상기 분석용 테이블형 데이터에 대한 분석 결과를 후속 처리 컴포넌트로 전달하는 단계를 포함하는, 방법을 제공한다. One embodiment provides a method performed by a computer, comprising: obtaining tabular data for analysis and storing the tabular data for analysis in at least one memory; encoding the tabular data for analysis using a first artificial intelligence model to generate latent variable data for analysis, wherein the first artificial intelligence model is pre-trained to generate latent variable data for analysis based on the tabular data for analysis and to restore segmented data for the tabular data for analysis based on the latent variable data for analysis; generating an analysis result for the tabular data for analysis using a second artificial intelligence model based on the latent variable data for analysis by the at least one processor; and transmitting the analysis result for the tabular data for analysis to a subsequent processing component.","assignee":"주식회사 Lg 경영개발원","inventors":["어문정","심예슬","임우형","조혜승","윤수희","윤상휴","이경은"],"publication_date":"2026-02-06","filing_date":"2026-01-29","priority_date":"2024-02-15","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260018141A/en"},{"publication_number":"KR20260018133A","title":"Method and apparatus for optimizing garment simulation parameters","abstract":"물성 파라미터 추정 방법 및 장치가 개시된다. 일 실시 예에 따른 가상 원단의 드레이프 시뮬레이션을 위한 물성 파라미터 추정 방법은 가상 원단에 대응하는 물성 파라미터를 신경망에 인가함으로써, 메쉬를 생성하는 단계, 메쉬에 기초하여, 가상 원단의 드레이프에 관한 타겟 데이터의 유형에 대응하는 드레이프 데이터를 획득하는 단계 및 획득된 드레이프 데이터와 타겟 데이터의 오차에 기초하여, 물성 파라미터 갱신하는 단계를 포함할 수 있다. A method and device for estimating material parameters are disclosed. According to one embodiment, a method for estimating material parameters for simulating a drape of a virtual fabric may include a step of generating a mesh by applying material parameters corresponding to the virtual fabric to a neural network, a step of obtaining drape data corresponding to a type of target data regarding a drape of the virtual fabric based on the mesh, and a step of updating the material parameters based on an error between the obtained drape data and the target data.","assignee":"(주)클로버추얼패션","inventors":["주은정","심응준","최명걸"],"publication_date":"2026-02-06","filing_date":"2026-01-27","priority_date":"2023-05-24","cpc_codes":["G","G06","G06T","G06T19/00","G06T19/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/10","G","G06","G06T","G06T17/00","G","G06","G06T","G06T17/00","G06T17/20","G","G06","G06T","G06T19/00","G","G06","G06T","G06T19/00","G06T19/003","G","G06","G06T","G06T3/00","G06T3/18","G","G06","G06T","G06T7/00","G06T7/97","G","G06","G06T","G06T2200/00","G06T2200/24","G","G06","G06T","G06T2210/00","G06T2210/16","G","G06","G06T","G06T2219/00","G06T2219/20","G06T2219/2021"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260018133A/en"},{"publication_number":"CN121481774A","title":"IoT-based intelligent building energy consumption monitoring system","abstract":"本申请涉及建筑能耗管理领域，具体公开了一种基于物联网的建筑能耗智能监测系统，其首先，通过实时感知能耗波动性、变化速率及与物理事件的耦合强度，智能地判断系统当前是处于平稳还是剧变状态，从而动态调整数据分析的时间窗口长度，实现对关键事件变焦观测。其次，在确定的自适应窗口内，它为每个数据点赋予一个随时间衰减的新鲜度权重，确保特征提取时能聚焦于最新的信息。通过这种方式，本方案将杂乱的异步数据流，重构为一个能精确表征系统当前动态特性的标准化特征向量，从根本上解决了因数据时间失配导致的特征失真问题，为上层AI模型的精准监测提供了高质量的数据基础。 This application relates to the field of building energy consumption management, specifically disclosing an IoT-based intelligent building energy consumption monitoring system. Firstly, by real-time sensing of energy consumption fluctuations, rates of change, and the coupling strength with physical events, it intelligently determines whether the system is currently in a stable or rapidly changing state, thereby dynamically adjusting the length of the data analysis time window to achieve focused observation of key events. Secondly, within a defined adaptive window, it assigns a freshness weight that decays over time to each data point, ensuring that feature extraction focuses on the latest information. In this way, this solution reconstructs the chaotic asynchronous data stream into a standardized feature vector that accurately represents the current dynamic characteristics of the system, fundamentally solving the feature distortion problem caused by data time mismatch and providing a high-quality data foundation for accurate monitoring by upper-level AI models.","assignee":"Zhejiang Xinyou Construction Engineering Co ltd","inventors":["许建成","唐平亚"],"publication_date":"2026-02-06","filing_date":"2026-01-14","priority_date":"2026-01-14","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/06","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06Q","G06Q50/00","G06Q50/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121481774A/en"},{"publication_number":"CN121469371A","title":"Intelligent Control Method and System for Charge Stack Power Module Based on Multi-Dimensional State Awareness","abstract":"The invention relates to the technical field of charging infrastructure electric energy management and power scheduling control, in particular to a charging pile power module intelligent control method and system based on multidimensional state awareness. The method comprises the steps of obtaining voltage and power constraint on a power grid side, current and temperature operation time of output voltage of a power module side, SOC and priority of power vehicles required by each gun on a load side, constructing a module state vector, calculating a capability evaluation value and a fatigue evaluation value, dividing the module into a high-capability group, a standard group and a limited group, dynamically grouping according to group priority under the power constraint, generating power distribution of each gun, determining target power supply to realize differentiated power supply if necessary, predicting load and constraint change by combining historical data, adjusting evaluation parameters and grouping priority to implement rotation, and issuing start and stop and a power set value to form closed-loop scheduling, so that the utilization rate is improved, and the over-temperature and fault risk are reduced. The system includes acquisition, evaluation grouping, grouping assignment, prediction adjustment, and execution modules.","assignee":"Wenzhou Blue Sky Energy Polytron Technologies Inc","inventors":["李伟伟","马永威","方介用","白益兵","张升腾","童仕鹏","刘汉庆"],"publication_date":"2026-02-06","filing_date":"2026-01-12","priority_date":"2026-01-12","cpc_codes":["B","B60","B60L","B60L53/00","B60L53/60","B60L53/62","G","G06","G06N","G06N20/00","Y","Y02","Y02T","Y02T10/00","Y02T10/60","Y02T10/7072","Y","Y02","Y02T","Y02T90/00","Y02T90/10","Y02T90/12"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121469371A/en"},{"publication_number":"KR20260017461A","title":"Method and apparatus for summarizing a document","abstract":"본 출원의 일 실시예에 따른 문서 요약 방법은, 사용자 검색 쿼리에 근거하여 요약 대상 문서를 추출하는 단계; 학습이 완료된 문서 요약 모델을 획득하는 단계; 및 학습이 완료된 문서 요약 모델에 상기 요약 대상 문서를 입력하고, 상기 학습이 완료된 문서 요약 모델의 제1 출력 레이어를 통하여 제1 요약 벡터를 획득하고, 상기 제1 요약 벡터에 기반하여 생성 요약 기반의 요약문을 획득하는 단계;를 포함하되, 상기 문서 요약 모델은, 문서 원본과 생성 요약 정답 정보로 구성된 학습 데이터 셋에 기반하여 훈련되되, 입력 레이어를 통하여 상기 문서 원본을 획득하고, 상기 문서 요약 모델의 제1 출력 레이어를 통하여 출력되는 출력 데이터와 상기 생성 요약 정답 정보의 차이에 기반하여 상기 생성 요약 정답 정보에 근사된 출력 데이터를 출력하도록 훈련된다. A document summarization method according to one embodiment of the present application comprises the steps of: extracting a document to be summarized based on a user search query; obtaining a document summary model for which training has been completed; and inputting the document to be summarized into the document summary model for which training has been completed, obtaining a first summary vector through a first output layer of the document summary model for which training has been completed, and obtaining a summary based on a generated summary based on the first summary vector; wherein the document summary model is trained based on a training data set comprising a document original and generated summary correct information, and is trained to obtain the document original through an input layer and output data approximated to the generated summary correct information based on a difference between output data output through the first output layer of the document summary model and the generated summary correct information.","assignee":"주식회사 포티투마루","inventors":["김동환","정우태","김현옥","박주식","장형진","손아림"],"publication_date":"2026-02-05","filing_date":"2026-01-26","priority_date":"2022-10-07","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/38","G06F16/383","G","G06","G06F","G06F16/00","G06F16/30","G06F16/34","G06F16/345","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2457","G06F16/24578","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3347","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F16/00","G06F16/90","G06F16/93","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260017461A/en"},{"publication_number":"AU2026200387A1","title":"Medical device location and tracking system","abstract":"Methods (500) and systems (300, 400) for automatically ascertaining physical location information about a plurality of medical device controllers (100) and tracking changes in the physical locations of ones of the respective medical device controllers. One or more machine learning modules (426, 430, 434) infer the physical locations from computer network messages received from the medical device controllers, including information about computer network components (306, 308, 310, 312) that are proximate ones of the medical device controllers (100).","assignee":"Abiomed Inc","inventors":["Alessandro Simone Agnello","Gregory John Eichmann","Paul Roland Lemay"],"publication_date":"2026-02-05","filing_date":"2026-01-20","priority_date":"2019-03-30","cpc_codes":["G","G05","G05B","G05B13/00","G05B13/02","G05B13/0265","G","G06","G06N","G06N20/00","G","G16","G16H","G16H40/00","G16H40/20","G","G16","G16H","G16H40/00","G16H40/60","G16H40/67"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026200387A1/en"},{"publication_number":"AU2026200343A1","title":"Generating images using sequences of generative neural networks","abstract":"Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating images. In one aspect, a method includes: receiving an input text prompt including a sequence of text tokens in a natural language; processing the input text prompt using a text encoder neural network to generate a set of contextual embeddings of the input text prompt; and processing the contextual embeddings through a sequence of generative neural networks to generate a final output image that depicts a scene that is described by the input text prompt.","assignee":"Google LLC","inventors":["William Chan","David James Fleet","Jonathan HO","Yi Li","Mohammad Norouzi","Chitwan SAHARIA","Saurabh Saxena","Jay Ha WHANG"],"publication_date":"2026-02-05","filing_date":"2026-01-19","priority_date":"2022-05-19","cpc_codes":["G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06F","G06F40/00","G06F40/40","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T11/00","G06T11/60","G","G06","G06T","G06T3/00","G06T3/40","G06T3/4053","G","G06","G06T","G06T5/00","G06T5/70","G","G06","G06T","G06T2211/00","G06T2211/40","G06T2211/441","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V30/00","G06V30/10","G06V30/19","G06V30/196","G06V30/1983"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026200343A1/en"},{"publication_number":"AU2026200328A1","title":"System and method for providing patient-specific dosing as a function of mathematical models updated to account for an observed patient response","abstract":"A system and method for predicting, proposing and/or evaluating suitable medication dosing regimens for a specific individual as a function of individual-specific characteristics and observed responses of the specific individual. Mathematical models of observed patient responses are used in determining an initial dose. The system and method use the patient’s observed response to the initial dose to refine the model for use to forecast expected responses to proposed dosing regimens more accurately for a specific patient. More specifically, the system and method uses Bayesian averaging, Bayesian updating and Bayesian forecasting techniques to develop patient-specific dosing regimens as a function of not only generic mathematical models and patient- specific characteristics accounted for in the models as covariate patient factors, but also observed patient-specific responses that are not accounted for within the models themselves, and that reflect variability that distinguishes the specific patient from the typical patient reflected by the model.","assignee":"Individual","inventors":["Diane Mould"],"publication_date":"2026-02-05","filing_date":"2026-01-16","priority_date":"2012-10-05","cpc_codes":["G","G01","G01N","G01N33/00","G01N33/48","G","G06","G06N","G06N7/00","G06N7/01","G","G16","G16H","G16H20/00","G16H20/10","G","G16","G16H","G16H50/00","G16H50/50"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026200328A1/en"},{"publication_number":"AU2026200278A1","title":"Allocating computing resources between model size and training data during training of a machine learning model","abstract":"A method performed by one or more computers, the method comprising: determining an energy budget for training a machine learning model to perform a machine learning task; determining a hardware specification defining a number of neural network accelerators available for training the machine learning model; determining a compute budget based on the energy budget and the hardware specification, wherein the compute budget characterizes an amount of computing resources allocated for training the machine learning model to perform the machine learning task; processing the data defining the compute budget using an allocation mapping, in accordance with a set of allocation mapping parameters, to generate an allocation tuple defining: (i) a target model size for the machine learning model, and (ii) a target amount of training data for training the machine learning model, wherein selecting a model size of the machine learning model as the target model size and training the machine learning model on the target amount of training data is predicted to optimize a performance of the machine learning model on the machine learning task subject to a constraint that an amount of computing resources used for training the machine learning model satisfies a threshold defined by the compute budget; instantiating the machine learning model, wherein the machine learning model has the target model size; obtaining the target amount of training data for training the machine learning model; and training the machine learning model having the target model size on the target amount of training data using hardware corresponding to the hardware specification.","assignee":"GDM Holding LLC","inventors":["Sebastian BORGEAUD DIT AVOCAT","Jordan HOFFMANN","Arthur MENSCH","Laurent Sifre"],"publication_date":"2026-02-05","filing_date":"2026-01-15","priority_date":"2022-03-29","cpc_codes":["G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5027","G06F9/505","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5011","G06F9/5016","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5027","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5027","G06F9/5044","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5094","G","G06","G06N","G06N20/00","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/501","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/5022","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/503","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/504","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/506"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026200278A1/en"},{"publication_number":"KR20260015970A","title":"Method and system for large language models alignment","abstract":"본 발명은 언어 모델 정렬 방법 및 시스템에 대한 것이다. 보다 구체적으로, 본 발명은 자기반성적 피드백(Self-Reflective Feedback)을 통한 언어 모델 정렬 방법 및 시스템에 대한 것이다. The present invention relates to a method and system for aligning language models. More specifically, the present invention relates to a method and system for aligning language models using self-reflective feedback.","assignee":"주식회사 Lg 경영개발원","inventors":["이경재","황다솔","박성현","장영수","이문태"],"publication_date":"2026-02-03","filing_date":"2026-01-20","priority_date":"2024-03-15","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260015970A/en"},{"publication_number":"DE202026100219U1","title":"AI-accelerated computing card for edge computing","abstract":"KI-beschleunigte Rechenkarte für Edge-Computing, dadurch gekennzeichnet, dass sie Folgendes umfasst:einen Chip der KI-beschleunigten Rechenkarte, der am Netzwerkrand angeordnet ist, wobei der Chip der KI-beschleunigten Rechenkarte in einem Schutzgehäuse gekapselt ist,wobei der Chip der KI-beschleunigten Rechenkarte eine auf einer Hauptplatine integrierte Referenzspannungsquelle SVG, einen ersten Leitungsschalter LineSwitch1 einer Transkonduktanzverstärkerschaltung TCA, einen Komparator COMP der Transkonduktanzverstärkerschaltung TCA, ein Register REG einer Sense-Verstärkerschaltung SAC, einen zweiten Leitungsschalter LineSwitch2 der Sense-Verstärkerschaltung SAC, einen dritten Leitungsschalter LineSwitch3 der Sense-Verstärkerschaltung SAC, Multiplikatoren MUX1, MUX3 der Sense-Verstärkerschaltung SAC und einen Zähler CNT umfasst. AI-accelerated computing card for edge computing, characterized in that it comprises: a chip of the AI-accelerated computing card located at the network edge, wherein the chip of the AI-accelerated computing card is encapsulated in a protective housing, the chip of the AI-accelerated computing card comprising a reference voltage source SVG integrated on a mainboard, a first line switch LineSwitch1 of a transconductance amplifier circuit TCA, a comparator COMP of the transconductance amplifier circuit TCA, a register REG of a sense amplifier circuit SAC, a second line switch LineSwitch2 of the sense amplifier circuit SAC, a third line switch LineSwitch3 of the sense amplifier circuit SAC, multipliers MUX1, MUX3 of the sense amplifier circuit SAC and a counter CNT.","assignee":"Individual","inventors":[],"publication_date":"2026-02-03","filing_date":"2026-01-15","priority_date":"2026-01-15","cpc_codes":["G","G06","G06F","G06F1/00","G06F1/16","G06F1/18","G06F1/183","G06F1/185","G","G06","G06F","G06F13/00","G06F13/38","G06F13/40","G06F13/4004","G06F13/4022","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063"],"country":"DE","kind":"application","source_url":"https://patents.google.com/patent/DE202026100219U1/en"},{"publication_number":"KR20260016578A","title":"Method for providing mesures for preventing onset of diabetes","abstract":"본 발명은 당뇨병의 예방을 위한 방안의 제공 방법에 관한 것으로, 적어도 2개의 주기에 걸친 복수의 대상자의 정보를 입력하는 단계, 여기서 상기 정보는 대상자별로의 당뇨병 발생 여부 및 의료변수에 대한 값을 포함한다; 상기 정보를 기초로 양방향 시계열 인공지능을 이용하여 예측모델을 형성하는 단계; 상기 예측모델에서 복수의 예측시점 별로 당뇨병의 발생을 예측하고 상기 의료변수 별로 가중치를 제공하는 단계 및 상기 의료변수 별 및 상기 예측시점 별로의 가중치에 기초하여 당뇨병 예방을 위한 방안을 제공하는 단계를 포함한다. The present invention relates to a method for providing a plan for preventing diabetes, comprising the steps of: inputting information of a plurality of subjects over at least two periods, wherein the information includes whether diabetes occurred and values for medical variables for each subject; forming a prediction model using two-way time series artificial intelligence based on the information; predicting the occurrence of diabetes for each of a plurality of prediction time points in the prediction model and providing weights for each of the medical variables; and providing a plan for preventing diabetes based on the weights for each of the medical variables and each of the prediction time points.","assignee":"한국수력원자력 주식회사","inventors":["김상돈","변정현","조성훈"],"publication_date":"2026-02-03","filing_date":"2026-01-14","priority_date":"2023-08-14","cpc_codes":["G","G16","G16H","G16H20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/50","G","G16","G16H","G16H50/00","G16H50/70"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260016578A/en"},{"publication_number":"KR20260015957A","title":"Personal Context-Aware Artificial Intelligence System for Determining Management Targets","abstract":"본 발명은 개인 종속 상황 인지 기반 인공지능 시스템에 관한 것으로서, 보다 구체적으로는 사용자의 상황 정보, 관심 대상 및 과거 이력에 기초하여 관리 대상 또는 관심 객체에 대한 판단 정보를 제공하는 인공지능 시스템에 관한 것이다. 본 발명에 따른 개인 종속 인공지능 시스템은 사용자의 상황 정보를 입력받아, 상기 상황 정보에 기초하여 관심 객체 또는 관리 대상을 식별하고, 식별된 관리 대상과 연관된 복수의 데이터 소스를 자동으로 선택하여 분석한다. 또한, 상기 분석 결과를 사용자에게 음성, 시각 또는 요약 정보 형태로 제공한다. 본 발명은 특정 산업, 특정 직업 또는 특정 하드웨어에 한정되지 않으며, 사용자가 동시에 수행하는 복수의 관리 역할과 관심 객체에 대해 상황 인지 기반의 판단 정보를 제공할 수 있다. 이를 통해 사용자는 별도의 복잡한 조작 없이도 자신의 현재 상황에 적합한 관리 대상 정보를 효율적으로 확인할 수 있다. The present invention relates to an artificial intelligence system based on personal contextual awareness, and more specifically, to an artificial intelligence system that provides judgment information on a management target or an object of interest based on a user's contextual information, object of interest, and past history. The personal AI system according to the present invention receives user contextual information, identifies objects of interest or management targets based on the contextual information, and automatically selects and analyzes multiple data sources associated with the identified management targets. Furthermore, the system provides the user with the results of the analysis in the form of audio, visual, or summary information. The present invention is not limited to a specific industry, occupation, or hardware, and can provide context-aware judgment information for multiple management roles and objects of interest simultaneously performed by users. This allows users to efficiently identify management target information appropriate to their current situation without the need for complex, separate operations.","assignee":"이호근","inventors":["이호근"],"publication_date":"2026-02-03","filing_date":"2026-01-12","priority_date":"2026-01-12","cpc_codes":["G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260015957A/en"},{"publication_number":"CN121458986A","title":"Tree image semantic matting method based on multi-scale dynamic attention","abstract":"The invention belongs to the technical field of image processing, and discloses a tree image semantic matting method based on multi-scale dynamic attention, which comprises the steps of preprocessing an original tree image and encoding target bounding box information as a spatial position embedding feature; the method comprises the steps of extracting multi-level semantic features of an image through a semantic perception coding module, fusing the multi-level semantic features with space position embedded features to form an enhanced multi-scale semantic feature set, inputting the features into a dynamic enhancement matting module, extracting and restoring the features by means of a U-Net backbone network, embedding a multi-scale dynamic space attention module in a bottleneck layer to adaptively enhance key features, simultaneously integrating an adaptive edge enhancement module in parallel in a decoding path to refine a complex edge structure, and finally fusing the multi-scale features to generate a high-quality transparency mask. According to the invention, through a cooperative mechanism of multi-scale dynamic attention and self-adaptive edge enhancement, the matting precision of complex edges and semitransparent areas in the tree image is effectively improved.","assignee":"Jiangxi Normal University","inventors":["郭凡","曾纪国","周德林","聂俊杰","汪婷","刘云骏","杨波"],"publication_date":"2026-02-03","filing_date":"2026-01-08","priority_date":"2026-01-08","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/20","G06V10/26","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121458986A/en"},{"publication_number":"CN121459364A","title":"Intelligent multimedia image recognition and automatic processing method and system","abstract":"本发明提供一种智能多介质图像识别与自动化处理方法及系统，方法包括：通过可见光摄像头获取包含目标介质的连续视频流；从视频流中抽取图像帧，利用旋转目标检测模型对图像帧进行实时解析，当检测到介质且其位置在连续多帧间保持稳定时，判定介质放置就绪，并对裁剪出的介质区域图像进行初步分类；根据初步分类结果，调用与该介质类型对应的专用识别流程，对介质区域图像进行处理，以获取结构化的识别结果数据；解析结构化的识别结果数据，根据其表征的介质类型与内容，生成并发送相应的控制指令，以驱动与终端设备连接的物理执行单元执行对应的自动化业务操作。本发明达到了硬件简化、全介质支持、流程自动化的技术效果。 This invention provides an intelligent multi-media image recognition and automated processing method and system. The method includes: acquiring a continuous video stream containing a target medium using a visible light camera; extracting image frames from the video stream; performing real-time analysis of the image frames using a rotating target detection model; determining that the medium is ready when it is detected and its position remains stable across multiple consecutive frames; performing preliminary classification of the cropped medium region image based on the preliminary classification result; calling a dedicated recognition process corresponding to the medium type to process the medium region image to obtain structured recognition result data; parsing the structured recognition result data; generating and sending corresponding control commands based on the medium type and content it represents to drive a physical execution unit connected to a terminal device to perform corresponding automated business operations. This invention achieves the technical effects of hardware simplification, full media support, and process automation.","assignee":"Shenzhen Sui Cai Technology Development Co ltd","inventors":["郭勇","袁森","刘可平"],"publication_date":"2026-02-03","filing_date":"2026-01-08","priority_date":"2026-01-08","cpc_codes":["G","G06","G06V","G06V30/00","G06V30/10","G06V30/18","G06V30/1801","G06V30/18019","G06V30/18038","G06V30/18048","G06V30/18057","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06V","G06V20/00","G06V20/40","G06V20/46","G","G06","G06V","G06V30/00","G06V30/10","G06V30/14","G06V30/148","G","G06","G06V","G06V30/00","G06V30/10","G06V30/19","G06V30/191","G06V30/19173"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121459364A/en"},{"publication_number":"CN121456123A","title":"Method and system for generating dialogue abstracts based on semantic structure fusion graph","abstract":"The invention provides a method and a system for generating a dialogue abstract based on a semantic structure fusion graph, which belong to the field of natural language processing and comprise the steps of constructing a dialogue intention feature graph according to intention types, constructing a time sequence feature dependency graph based on relative round distance, constructing an interaction frequency matrix of a dialogue user, expanding edge weights among dialogue nodes into the dialogue user interaction graph based on the interaction frequency matrix, constructing a self-learning attention mechanism graph based on a self-learning attention mechanism, fusing the dialogue intention feature graph, the time sequence feature dependency graph and the dialogue user interaction graph to generate a primary semantic fusion graph, inputting the self-learning attention mechanism graph into an input layer of a gating fusion encoder, inputting the primary semantic fusion graph and the self-learning attention mechanism graph into a gating unit of the gating fusion encoder to obtain coding features, and inputting the coding features into a decoder to generate a abstract of a dialogue text. The invention realizes the organic fusion of the explicit structure information and the implicit generation stage.","assignee":"Shandong Jiaotong University","inventors":["朱振方","王富瑞","高志修","张一鸣","付振睿","裴洪丽","卢强","赵大伟","亓江涛","徐泽西"],"publication_date":"2026-02-03","filing_date":"2026-01-08","priority_date":"2026-01-08","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/34","G06F16/345","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G06F40/211","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06F","G06F40/00","G06F40/30","G06F40/35","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","Y","Y02","Y02D","Y02D10/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121456123A/en"},{"publication_number":"CN121454355A","title":"Battery fault diagnosis method, device and system based on artificial intelligence","abstract":"The invention provides a battery fault diagnosis method, device and system based on artificial intelligence, which relate to the technical field of artificial intelligence, develop cross-feature convolution fusion based on the working condition mode of determining each attribute monitoring data, deeply mine multi-attribute cooperative change relation, superimpose and strengthen early weak fault features scattered in different channels, filter invalid interference caused by working condition fluctuation, avoid the weak fault features from being submerged by normal fluctuation and noise, and greatly improve early fault recognition capability. Meanwhile, on the premise of performing disturbance suppression on the cross-feature enhancement vector, a global perception feature representation with multi-attribute cooperative features and full-period evolution features is constructed, so that the progressive evolution of the battery fault can be quantitatively evaluated, the diagnosis model can be combined with the fault development trend to perform fault pre-judgment in advance, and fault deterioration is delayed or prevented from the source.","assignee":"Qingdao Tieqi Network Technology Co ltd","inventors":["陈宗兴","卢世亮","夏立国","杨珍花"],"publication_date":"2026-02-03","filing_date":"2026-01-08","priority_date":"2026-01-08","cpc_codes":["G","G01","G01R","G01R31/00","G01R31/36","G01R31/367","G","G01","G01R","G01R31/00","G01R31/36","G01R31/382","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2131","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2137","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121454355A/en"},{"publication_number":"KR20260015324A","title":"Method and system for processing task in parallel","abstract":"본 개시는 적어도 하나의 프로세서에 의해 수행되는 태스크 병렬 처리 방법이 제공된다. 이 방법은, 제1 인스트럭션(instruction)과 연관된 제1 태스크(task)를 수행하는 단계, 제1 인스트럭션이 버스트 로드 인스트럭션인지 여부를 판정하는 단계, 제1 인스트럭션이 버스트 로드 인스트럭션으로 판정된 것에 응답하여, 제2 인스트럭션을 획득하는 단계 및 획득된 제2 인스트럭션과 연관된 제2 태스크를 수행하는 단계를 포함하고, 제1 태스크와 제2 태스크는 병렬적으로 수행될 수 있다. The present disclosure provides a method for task parallel processing performed by at least one processor. The method comprises the steps of performing a first task associated with a first instruction, determining whether the first instruction is a burst load instruction, obtaining a second instruction in response to determining that the first instruction is a burst load instruction, and performing a second task associated with the obtained second instruction, wherein the first task and the second task can be performed in parallel.","assignee":"리벨리온 주식회사","inventors":["김현호","김진석","오진욱"],"publication_date":"2026-02-02","filing_date":"2026-01-21","priority_date":"2023-03-20","cpc_codes":["G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/38","G06F9/3885","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5083","G","G06","G06F","G06F15/00","G06F15/76","G06F15/78","G06F15/7807","G","G06","G06F","G06F15/00","G06F15/76","G06F15/78","G06F15/7828","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/30003","G06F9/3004","G06F9/30043","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/38","G06F9/3824","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/38","G06F9/3824","G06F9/3834","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/38","G06F9/3836","G06F9/3851","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/38","G06F9/3854","G06F9/3856","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/38","G06F9/3867","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/48","G06F9/4806","G06F9/4812","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/48","G06F9/4806","G06F9/4843","G06F9/4881","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260015324A/en"},{"publication_number":"KR20260014683A","title":"Multimodal Emotion Weight Fusion-based Automatic Video Segment Classification and Generative AI Parameter Linked Control System","abstract":"본 발명은 멀티모달 감정 가중치 융합 기반의 영상 구간 자동 분류 및 생성형 AI 파라미터 연동 제어 시스템에 관한 것으로, 입력 영상의 비디오 채널에서 안면 랜드마크를 추출하여 VAD(Valence-Arousal-Dominance) 3차원 연속 공간의 시각 감정 벡터(V_emo ∈ ?³)를 생성하고, 오디오 채널에서 멜-주파수 케프스트럼 계수(MFCC)를 추출하여 동일한 VAD 공간의 청각 감정 벡터(A_emo ∈ ?³)를 생성한 후, 두 벡터 간의 코사인 유사도를 상관관계 지수(ρ)로 계산하고, 환경 노이즈 레벨(N_env) 및 각 모달리티의 인식 신뢰도(C_v, C_a)를 입력으로 하는 신경망 기반 함수 f(ρ, N_env, C_v, C_a)를 통해 적응형 신뢰도 가중치(α)를 산출하며, 이를 적용하여 통합 감정 지수(E_score = α · V_emo + (1-α) · A_emo)를 연산하고, 지수 가중 이동평균 필터로 평활화된 감정 곡선 E_smooth(t)를 생성한다. 또한, 평활화된 감정 곡선의 1차 미분값(dE/dt)과 2차 미분값(d²E/dt²)을 분석하여 감정 전환점을 정밀 검출하고, 감정 전환점 사이의 구간을 고유 감정 이벤트 구간으로 정의하여 숏폼 콘텐츠 제작을 위한 컷 편집 지점을 자동으로 결정하며, VAD 좌표를 HSL 색공간에 매핑하는 F_color, BPM에 매핑하는 F_tempo, 폰트 스타일에 매핑하는 F_font 변환 함수를 통해 영상 효과를 감정에 동기화하여 자동 제어한다. The present invention relates to a system for automatically classifying video sections and controlling generative AI parameters based on multimodal emotion weight fusion, wherein a facial landmark is extracted from a video channel of an input video to generate a visual emotion vector (V_emo ∈ ?³) in a valence-arousal-dominance (VAD) three-dimensional continuous space, and a mel-frequency Keppstrum coefficient (MFCC) is extracted from an audio channel to generate an auditory emotion vector (A_emo ∈ ?³) in the same VAD space, and then the cosine similarity between the two vectors is calculated as a correlation index (ρ), and an adaptive confidence weight (α) is calculated through a neural network-based function f(ρ, N_env, C_v, C_a) that takes as input an environmental noise level (N_env) and a recognition confidence (C_v, C_a) of each modality, and by applying this, an integrated emotion score (E_score = α · V_emo + (1-α) · A_emo) is calculated, and an exponentially weighted moving average is calculated. It generates an emotional curve E_smooth(t) smoothed by a filter. In addition, it precisely detects emotional turning points by analyzing the first derivative (dE/dt) and second derivative (d²E/dt²) of the smoothed emotional curve, and automatically determines cut editing points for short-form content production by defining the section between emotional turning points as a unique emotional event section. In addition, it automatically controls video effects by synchronizing them with emotions through the F_color conversion function that maps VAD coordinates to the HSL color space, F_tempo conversion function that maps them to BPM, and F_font conversion function that maps them to font styles.","assignee":"나형석","inventors":["나형석"],"publication_date":"2026-01-30","filing_date":"2026-01-13","priority_date":"2026-01-13","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06T","G06T3/00","G06T3/40","G06T3/4015","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G","G10","G10L","G10L25/00","G10L25/03","G10L25/24","G","G10","G10L","G10L25/00","G10L25/48","G10L25/51","G10L25/63"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260014683A/en"},{"publication_number":"CN121431543A","title":"A Precise Detection Method for Power Line Hazards Based on Intelligent Analysis of UAV Multispectral Collaborative Imaging","abstract":"本发明涉及基于无人机多光谱协同成像智能分析的电力线路隐患精准检测方法，包括规划无人机的巡检飞行路径，通过无人机对架空电力线路巡检对象进行多光谱图像采集；通过时空对齐法构建时空对齐变换矩阵；对采集的多光谱图像进行时空对齐；构建融合多光谱图像信息的端到端深度学习网络，并将预处理后的多光谱图像输入至端到端深度学习网络，得到电力线路隐患的检测结果。本发明通过融合可见光、红外、紫外等多模态特征，突破单光谱或简单拼接多光谱的信息利用局限，让缺陷隐患检测模型更全面感知隐患。充分整合了多光谱数据的优势，实现对架空电力线路隐患的精准检测，提高了检测的准确率和可靠性，更及时、准确地发现电力线路存在的隐患问题。 This invention relates to a method for accurate detection of power line hazards based on intelligent analysis of multispectral collaborative imaging using unmanned aerial vehicles (UAVs). The method includes planning the UAV's inspection flight path, acquiring multispectral images of overhead power lines using the UAV, constructing a spatiotemporal alignment transformation matrix using a spatiotemporal alignment method, performing spatiotemporal alignment on the acquired multispectral images, constructing an end-to-end deep learning network that fuses multispectral image information, and inputting the preprocessed multispectral images into the end-to-end deep learning network to obtain the detection results of power line hazards. This invention overcomes the limitations of single-spectral or simply stitched multispectral information utilization by fusing visible light, infrared, and ultraviolet multimodal features, allowing the defect and hazard detection model to more comprehensively perceive hazards. It fully integrates the advantages of multispectral data, achieving accurate detection of hazards in overhead power lines, improving detection accuracy and reliability, and enabling more timely and accurate discovery of potential hazards in power lines.","assignee":"Zhilian Xinneng Power Technology Co ltd; China Electric Power Research Institute Co Ltd CEPRI; State Grid East Inner Mongolia Electric Power Co Ltd","inventors":["刘壮","付家兴","谈发力","刘厚轩","陈霖然","蔡焕青","付晶","曹宇钊","殷志江","邵瑰玮"],"publication_date":"2026-01-30","filing_date":"2025-12-31","priority_date":"2025-12-31","cpc_codes":["G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/95","G01N21/952","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/10","G06V20/17","H","H02","H02G","H02G1/00","H02G1/02","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G01N2021/8887"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121431543A/en"},{"publication_number":"CN121432944A","title":"Intelligent control methods for regional forestry pests","abstract":"本发明为林业保护技术领域，具体为区域林业有害生物智能防控方法，该方法包括四步流程：通过智能测报站拍摄虫情图像及采集环境因子数据，无人机基于A星算法生成巡航路径，结合多光谱图像与克里金插值生成虫害密度热力图；建立害虫特征模板库，通过切比雪夫距离和余弦相似度筛选候选种类，再经特征加权投票识别虫害类型；基于熵权法与物元可拓算法构建模型，融合环境适宜关联度与林业密度风险值，得到虫害风险关联度并预警；通过规则与强化学习生成杀虫灯最优功率控制指令，采用遗传算法获取无人机最优喷药路径。该方法实现虫害监测、识别、预警与防控的全流程智能化，适配林区防控高效的实际需求。 This invention pertains to the field of forestry protection technology, specifically a method for intelligent control of regional forestry pests. The method comprises four steps: 1) Capturing pest images and collecting environmental factor data through intelligent monitoring stations; 2) Generating a flight path for a drone based on the A*Sat algorithm; 3) Generating a pest density heatmap by combining multispectral images with Kriging interpolation; 4) Establishing a pest feature template library; 5) Screening candidate species using Chebyshev distance and cosine similarity; 6) Identifying pest types through feature-weighted voting; 7) Constructing a model based on entropy weighting and matter-element extension algorithms; 8) Integrating environmental suitability correlation and forestry density risk values to obtain pest risk correlation and provide early warning; 9) Generating optimal power control commands for insecticidal lamps through rule-based and reinforcement learning; and 10) Obtaining the optimal spraying path for the drone using a genetic algorithm. This method achieves intelligent control of the entire process of pest monitoring, identification, early warning, and control, meeting the practical needs of efficient pest control in forest areas.","assignee":"Hunan Linkoda Information Technology Co ltd; Hunan Linkeda Agriculture And Forestry Technical Service Co ltd","inventors":["伍南","周刚","张烜","赵正萍","李青","赵娜"],"publication_date":"2026-01-30","filing_date":"2025-12-31","priority_date":"2025-12-31","cpc_codes":["G","G05","G05B","G05B13/00","G05B13/02","G05B13/0265","A","A01","A01M","A01M1/00","A01M1/02","A01M1/04","A","A01","A01M","A01M7/00","A01M7/0089","G","G01","G01C","G01C21/00","G01C21/20","G","G05","G05B","G05B13/00","G05B13/02","G05B13/04","G05B13/042","G","G05","G05D","G05D1/00","G05D1/60","G05D1/646","G","G05","G05D","G05D1/00","G05D1/60","G05D1/648","G05D1/6484","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/12","G06N3/126","G","G06","G06V","G06V10/00","G06V10/20","G06V10/26","G06V10/267","G","G06","G06V","G06V10/00","G06V10/40","G06V10/54","G","G06","G06V","G06V10/00","G06V10/40","G06V10/56","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/761","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/10","G06V20/17","G","G06","G06V","G06V40/00","G06V40/10"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121432944A/en"},{"publication_number":"CN121437514A","title":"A saliency enhancement adaptive cooperative sensing method for surface defect detection","abstract":"本发明提供了一种用于表面缺陷检测的显著性增强自适应协同感知方法，包括以下步骤：S100：构建表面缺陷数据集；S200：构建显著性增强自适应协同感知网络；S300：对显著性增强自适应协同感知网络进行训练；S400：利用训练完成的网络模型进行缺陷分割；以适应表面缺陷存在边缘模糊、尺度差异大、分布随机及光照条件复杂等情况，与采用传统卷积或单一路径特征提取的现有方案不同，本发明通过浅层显著性增强、高层多尺度语义建模以及解码阶段的边缘‑区域协同机制，使缺陷特征能够在不同尺度和不同结构层面得到有效表达，从而提升整体分割的准确性和稳定性。 This invention provides a saliency enhancement adaptive collaborative perception method for surface defect detection, comprising the following steps: S100: constructing a surface defect dataset; S200: constructing a saliency enhancement adaptive collaborative perception network; S300: training the saliency enhancement adaptive collaborative perception network; S400: performing defect segmentation using the trained network model. This method adapts to situations where surface defects exhibit blurred edges, large scale differences, random distribution, and complex lighting conditions. Unlike existing schemes that use traditional convolution or single-path feature extraction, this invention employs shallow saliency enhancement, high-level multi-scale semantic modeling, and an edge-region collaborative mechanism in the decoding stage. This enables defect features to be effectively expressed at different scales and structural levels, thereby improving the overall accuracy and stability of segmentation.","assignee":"Hrlm Technology Inc Co","inventors":["丁传仓","李前程","王报祥","魏勇","江星星","唐春","高庆辉","杨强","陈茜茜"],"publication_date":"2026-01-30","filing_date":"2025-12-31","priority_date":"2025-12-31","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T7/00","G06T7/10","G06T7/12","G","G06","G06T","G06T7/00","G06T7/10","G06T7/13","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","Y","Y02","Y02P","Y02P90/00","Y02P90/30"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121437514A/en"},{"publication_number":"CN121433764A","title":"A method and system for automatically running operation and maintenance tasks","abstract":"本发明公开了一种运维任务自动运行的方法及系统；脚本运维环境依赖风险高及复杂任务调度低效，本发明通过获取低代码编排特征、Playbook结构特征以及主机负载参数合并融合得到自动运维特征，并获取服务器证书执行方法，利用自动运维特征和服务器证书执行方法构建证书执行运维自动化模型。本发明考虑低代码编排特征、Playbook结构特征以及主机负载参数来构建多参数修正神经网络模型，通过强化学习方法生成更多Playbook结构特征数据，使得证书执行运维自动化模型更加精准智能地得出服务器证书智能执行方法，同时任务在容器沙箱中运行以隔离风险，结合拖拽式流程图参数生成可执行脚本的自愈机制，应用于多服务器证书更新时智能分批执行并规避故障节点。 This invention discloses a method and system for automatically running operation and maintenance tasks. Addressing the high risks and inefficient scheduling of complex tasks in script-based operation and maintenance environments, this invention obtains automatic operation and maintenance features by acquiring and fusing low-code orchestration features, Playbook structure features, and host load parameters. It also acquires server certificate execution methods and constructs an automated certificate execution operation and maintenance model using these features and methods. This invention considers low-code orchestration features, Playbook structure features, and host load parameters to construct a multi-parameter corrected neural network model. Reinforcement learning methods generate more Playbook structure feature data, enabling the automated certificate execution operation and maintenance model to more accurately and intelligently derive intelligent server certificate execution methods. Simultaneously, tasks run in a container sandbox to isolate risks. Combined with a drag-and-drop flowchart parameter generation executable script self-healing mechanism, this is applied to intelligently batch execution and avoidance of faulty nodes during multi-server certificate updates.","assignee":"Guangzhou Shanghang Information Technology Co ltd","inventors":["兰满桔","刘杰","赵伟锋"],"publication_date":"2026-01-30","filing_date":"2025-12-31","priority_date":"2025-12-31","cpc_codes":["G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/448","G06F9/4482","G","G06","G06F","G06F21/00","G06F21/70","G06F21/71","G06F21/74","G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/455","G06F9/45533","G06F9/45558","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/455","G06F9/45533","G06F9/45558","G06F2009/45587"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121433764A/en"},{"publication_number":"CN121434793A","title":"A multi-agent-based data augmentation method and apparatus","abstract":"本申请公开了一种基于多智能体的数据增强方法及装置，该方法包括：规划智能体获取样本数据的任务描述以及数据描述，数据描述包括数据分布统计和数据特征明细；规划智能体依照样本数据的数据分布统计和数据特征明细，生成初始的零样本思维链，零样本思维链包括每个执行层级调用的智能体类型和执行策略；规划智能体基于零样本思维链的执行策略，确定调用增强策略智能体的执行层级，生成新样本；规划智能体评估新样本对下游的深度学习模型的增强效果，若增强效果未达到设定条件，则调整零样本思维链的执行层级顺序和执行层级中的数据增强步骤，并返回基于零样本思维链确定调用增强策略智能体的执行层级的步骤，以提升深度学习模型效率。 This application discloses a data augmentation method and apparatus based on multi-agent systems. The method includes: planning a task description and data description for an agent to acquire sample data, the data description including data distribution statistics and data feature details; the planning agent generating an initial zero-shot thought chain based on the data distribution statistics and data feature details of the sample data, the zero-shot thought chain including the agent type and execution strategy called at each execution level; the planning agent determining the execution level of the agent calling the augmentation strategy based on the execution strategy of the zero-shot thought chain, and generating new samples; the planning agent evaluating the augmentation effect of the new samples on the downstream deep learning model, if the augmentation effect does not meet the set conditions, adjusting the execution level order and data augmentation steps in the execution level of the zero-shot thought chain, and returning to the step of determining the execution level of the agent calling the augmentation strategy based on the zero-shot thought chain, so as to improve the efficiency of the deep learning model.","assignee":"Hangzhou Hikvision Digital Technology Co Ltd","inventors":["段路云","邱星"],"publication_date":"2026-01-30","filing_date":"2025-12-29","priority_date":"2025-12-29","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F18/00","G06F18/20","G06F18/27","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121434793A/en"},{"publication_number":"CN121437496A","title":"A Closed-Loop Monitoring Method and System for Planar Sorting Process Based on Visual Perception and Reasoning","abstract":"The application relates to a plane sorting flow closed-loop monitoring method and system based on visual perception and reasoning, wherein the method comprises the steps of obtaining overlooking images covering a plurality of working areas on a sorting plane, carrying out affine transformation on area images corresponding to the working areas in the overlooking images respectively to obtain standard visual angle images of the working areas, adopting a shared visual recognition model to recognize the standard visual angle images, outputting the state and material information of a mechanical arm in each working area, constructing a time sequence event stream which changes the mechanical arm action and the area material into events based on the recognition result of continuous multiframes, carrying out real-time compliance diagnosis on the time sequence event stream according to a preset flow logic rule base to detect flow abnormality, and executing corresponding grading response and flow state recovery operation according to the severity level of the abnormality when the flow abnormality is detected. The application realizes the non-blind area state sensing of the plane sorting flow, improves the monitoring efficiency of the plane sorting flow and reduces the cost.","assignee":"Changzhou Shiyuan Technology Co ltd","inventors":["郭晓觅","杨浩","吴雨豪","李威剑","尹天骄","童毅炜","张逸颖","樊天宇"],"publication_date":"2026-01-30","filing_date":"2025-12-19","priority_date":"2025-12-19","cpc_codes":["B","B07","B07C","B07C5/00","B07C5/34","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G","G06","G06V","G06V10/00","G06V10/20","G06V10/24","G06V10/247","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V20/00","G06V20/50","G06V20/52","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30108","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30232","G","G06","G06V","G06V2201/00","G06V2201/06"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121437496A/en"},{"publication_number":"AU2026200121A1","title":"Prediction of usage or compliance","abstract":"22351536_1 (GHMatters) P116641.AU.1 Systems and methods have been developed to increase user compliance and adherence to various devices and services including respiratory therapy devices, exercise equipment, and online or other software services. For instance, in some examples the disclosed technology may monitor usage data output from a respiratory therapy device, exercise equipment or computer software program to determine, based on the trends of usage, when a user is likely to terminate or reduce usage within a specified time window. Flagging a user may also trigger further actions to automatically intervene before the user terminates engagement with the service.","assignee":"Resmed Inc","inventors":["Sakeena DE SOUZA","Oleksandr GROMENKO","Nathan Liu"],"publication_date":"2026-01-29","filing_date":"2026-01-08","priority_date":"2018-12-28","cpc_codes":["A","A61","A61M","A61M16/00","A61M16/021","A61M16/022","A61M16/024","A61M16/026","G","G16","G16H","G16H40/00","G16H40/60","G16H40/67","G","G06","G06F","G06F21/00","G06F21/30","G06F21/31","A","A61","A61B","A61B5/00","A61B5/08","A61B5/087","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4833","A","A61","A61M","A61M16/00","A61M16/0003","A","A61","A61M","A61M16/00","A61M16/0051","A","A61","A61M","A61M16/00","A61M16/0057","A61M16/0066","A","A61","A61M","A61M16/00","A61M16/06","A","A61","A61M","A61M16/00","A61M16/06","A61M16/0605","A61M16/0616","G","G06","G06N","G06N20/00","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H20/00","G16H20/30","G","G16","G16H","G16H20/00","G16H20/40","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/70","G","G16","G16H","G16H80/00","A","A61","A61M","A61M16/00","A61M16/10","A61M16/14","A61M16/16","A","A61","A61M","A61M16/00","A61M16/0003","A61M2016/0027","A","A61","A61M","A61M16/00","A61M16/0003","A61M2016/003","A","A61","A61M","A61M2202/00","A61M2202/02","A61M2202/0208","A","A61","A61M"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2026200121A1/en"},{"publication_number":"KR20260014021A","title":"Direction Estimation Apparatus of Power Flow in case of Failure","abstract":"본 발명의 고장시 전력 조류의 방향 추정 장치는, 배전 계통의 대상 계측지점에서의 순시치 전류 파형을 모니터링하는 모니터링부; 상기 계측지점의 전류 크기가 소정의 기준값을 넘으면 고장으로 판정하는 고장 판정부; 모니터링하는 상기 순시치 전류 파형의 순시치 전류 패턴을 수집하는 전류 패턴 수집부; 및 고장으로 판정한 픽업 시점 이전의 0교차전 시점의 순시치 전류 극성과 상기 픽업 시점 직후 순시치 전류의 극성이 동일하면 고장 조류의 방향을 정상 조류의 방향과 반대로 추정하고, 극성들이 서로 다르면 고장 조류의 방향을 정상 조류의 방향과 동일로 추정하는 조류 방향 추정부를 포함할 수 있다. The device for estimating the direction of power flow in case of a fault of the present invention may include: a monitoring unit that monitors an instantaneous current waveform at a target measuring point of a power distribution system; a fault determining unit that determines a fault when the current magnitude of the measuring point exceeds a predetermined reference value; a current pattern collecting unit that collects an instantaneous current pattern of the monitored instantaneous current waveform; and a current direction estimating unit that estimates the direction of the fault current as opposite to the direction of the normal current when the polarity of the instantaneous current at a time point before the zero crossing before the pick-up time point determined to be a fault is the same as the polarity of the instantaneous current immediately after the pick-up time point, and estimates the direction of the fault current as the same as the direction of the normal current when the polarities are different.","assignee":"한국전력공사","inventors":["김우현","김주용","채우규","이현명","이춘권"],"publication_date":"2026-01-29","filing_date":"2026-01-07","priority_date":"2022-10-28","cpc_codes":["G","G01","G01R","G01R19/00","G01R19/14","G","G01","G01R","G01R19/00","G01R19/165","G01R19/16566","G01R19/16571","G","G01","G01R","G01R19/00","G01R19/165","G01R19/16566","G01R19/1659","G","G01","G01R","G01R22/00","G01R22/06","G01R22/061","G01R22/068","G","G01","G01R","G01R23/00","G01R23/16","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260014021A/en"},{"publication_number":"KR20260013504A","title":"Defect inspection system using 3d measuring machine and defect inspection method using 3d measuring machine","abstract":"3D 측정기를 이용한 양불 검사 방법이 개시된다. 본 발명의 일 실시 예에 따른 3D 측정기를 이용한 양불 검사 방법은, 2D 이미지를 생성하는 단계(S11); 상기 2D 이미지로부터 소자의 형상정보를 추출하는 단계(S12); 상기 형상정보에 기초하여, 복수의 3D 측정 방식 중 적어도 하나를 선택하는 단계(S13); 및 선택된 3D 측정 방식에 따라 3D 이미지를 생성하는 단계(S14)를 포함할 수 있다. 3D 측정기를 이용한 양불 검사 시스템이 개시된다. 본 발명의 일 실시 예에 따른 3D 측정기를 이용한 양불 검사 시스템은, 로딩부와 언로딩부 사이에 개재되고, 소자를 실장하여 이송하는 이송부; 상기 이송부에 대향하여 마련되고, 상기 소자에 대한 2D 이미지를 생성하는 2D 측정기; 상기 2D 측정기와 나란하게 상기 이송부에 대향하여 마련되고, 상기 소자에 대한 3D 이미지를 생성 가능한 서로 다른 복수대의 3D 측정기; 상기 3D 이미지와 상기 2D 이미지를 정합 후 병합하여 소자 이미지를 생성하고, 상기 소자 이미지를 토대로 상기 소자의 불량 여부를 판정하는 프로세서; 및 상기 2D 이미지와 3D 이미지, 소자 이미지를 저장 가능한 메모리를 포함하고. 상기 프로세서는, 상기 2D 이미지로부터 상기 소자에 대한 형상정보들을 추출해, 상기 3D 측정기 중 어느 하나를 선택적으로 구동 제어 가능한 측정기 구동제어 유닛; 상기 소자 이미지를 보정하는 이미지 전처리 유닛; 기 저장된 소자 설계 정보를 토대로, 상기 소자 이미지에 대응되는 설계 이미지를 선별, 호출하는 이미지 선별 유닛; 및 상기 소자 이미지와 상기 설계 이미지를 비교 대조하는 양불 판정 유닛을 포함하고, 상기 메모리는, 상기 소자의 종류와 특성별로 소자 설계 정보 및 설계 이미지를 저장하고, 상기 이미지 전처리 유닛에 의해 보정 처리된 소자 이미지를 저장할 수 있다. A method for inspecting defects using a 3D measuring device is disclosed. According to an embodiment of the present invention, the method for inspecting defects using a 3D measuring device may include the steps of: generating a 2D image (S11); extracting shape information of a component from the 2D image (S12); selecting at least one of a plurality of 3D measuring methods based on the shape information (S13); and generating a 3D image according to the selected 3D measuring method (S14). A quality inspection system using a 3D measuring device is disclosed. According to an embodiment of the present invention, a quality inspection system using a 3D measuring device includes: a transport unit interposed between a loading unit and an unloading unit, for mounting and transporting a device; a 2D measuring device provided opposite the transport unit and generating a 2D image of the device; a plurality of different 3D measuring devices provided parallel to the 2D measuring devices and opposite the transport unit, the 3D measuring devices capable of generating a 3D image of the device; a processor for generating a device image by aligning and merging the 3D image and the 2D image, and determining whether the device is defective based on the device image; and a memory capable of storing the 2D image, the 3D image, and the device image. The processor comprises: a measuring device drive control unit capable of selectively driving and controlling one of the 3D measuring devices by extracting shape information of the device from the 2D image; an image preprocessing unit for correcting the device image; an image selection unit for selecting and calling a design image corresponding to the device image based on previously stored device design information; And it includes a pass/fail judgment unit that compares and contrasts the element image and the design image, and the memory can store element design information and design images according to the type and characteristics of the element, and store element images corrected by the image preprocessing unit.","assignee":"임영한","inventors":["임영한"],"publication_date":"2026-01-28","filing_date":"2026-01-19","priority_date":"2022-05-09","cpc_codes":["G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G","G01","G01B","G01B11/00","G01B11/24","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8806","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/95","G01N21/956","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/95","G01N21/956","G01N21/95607","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/95","G01N21/956","G01N21/95684","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T17/00","G06T17/30","G","G06","G06T","G06T5/00","G06T5/90","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G06T7/001","G","G06","G06T","G06T7/00","G06T7/30","G","G01","G01N","G01N21/00","G01N21/17","G01N2021/178","G01N2021/1785","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G01N2021/8854","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G01N2021/8883","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G01N2021/8887","G","G01","G01N","G01N2201/00","G01N2201/10","G01N2201/104","G01N2201/1042","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10028","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30108"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260013504A/en"},{"publication_number":"CN121414382A","title":"A method and system for determining feed compliance based on processing equipment information","abstract":"The invention belongs to the field of feed processing, and relates to a method and a system for determining feed mixing regulation degree based on processing equipment information, wherein the method comprises the steps of obtaining feed information and obtaining processing procedure information of the feed according to the feed information; the method comprises the steps of constructing initial process characteristics, cross characteristics and initial object characteristics based on processing process information, processing the initial process characteristics based on a processing equipment operation parameter library to obtain optimized process characteristics, processing the initial object characteristics based on a raw material information library to obtain optimized object characteristics, processing the optimized process characteristics, the cross characteristics and the optimized object characteristics to obtain verification characteristics, determining the compliance degree of feed according to the similarity between the verification characteristics and standard characteristics, and fusing feed compliance degree judging technologies of raw material characteristics, process association, dynamic environments and equipment states to realize accurate, efficient and traceable judgment of compliance and provide technical support for feed quality safety control.","assignee":"Sichuan Xinte Agriculture And Animal Husbandry Technology Co ltd","inventors":["袁庆刚","付通久","陈浩","谢峰","苑一哲"],"publication_date":"2026-01-27","filing_date":"2025-12-30","priority_date":"2025-12-30","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/018","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2433","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049","G","G06","G06Q","G06Q50/00","G06Q50/02","Y","Y02","Y02P","Y02P90/00","Y02P90/30"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121414382A/en"},{"publication_number":"CN121412599A","title":"A smart mating and extraction life prediction system and method for electronic connectors","abstract":"The application provides an intelligent plugging life prediction system and method for an electronic connector, which are used for respectively extracting time domain and frequency domain characteristics after preprocessing multidimensional physical signals collected by plugging operation each time so as to construct high-dimensional original characteristic vectors of the plugging operation each time, carrying out degradation evaluation on health states of the electronic connector in the plugging operation each time through the high-dimensional original characteristic vectors, outputting health indexes representing the current mechanical and electrical contact states of the electronic connector, further establishing a health index sequence of the electronic connector in the process of multiple plugging, carrying out trend extrapolation on future health states of the electronic connector based on attenuation trend in the health index sequence, and taking the difference between corresponding predicted plugging times and the current executed plugging times as the residual plugging life of the electronic connector when the predicted health index value is lower than a preset failure threshold for the first time. Based on the scheme, the health degree prediction based on multidimensional physical signal fusion modeling can be realized.","assignee":"Dongguan City Fine Precision Electronics Technology Co ltd","inventors":["孙超","陈剑锋","詹成扬","王玲平"],"publication_date":"2026-01-27","filing_date":"2025-12-30","priority_date":"2025-12-30","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121412599A/en"},{"publication_number":"CN121415780A","title":"Multi-module collaborative flight simulator control panel and method","abstract":"本发明属于航空飞行训练技术领域，具体涉及了一多模块协同的飞行模拟机操纵面板及方法，解决现有系统交互效率低、评估主观、预警滞后、讲评耗时的问题。本发明依托传统IOS硬件，并设置：智能语音交互模块通过ASR识别教员语音，经意图识别与三级槽位补全生成指令，搭配双层飞行阶段识别逻辑保障交互精准；飞行质量监控模块并行运行规则引擎与增强LSTM神经网络，结合自适应融合算法输出综合飞行质量分数；实时提醒模块按事件等级与飞行阶段动态调整预警优先级，提供多模式预警；智能讲评模块融合多源数据，按模板优先级生成个性化结构化报告。本发明实现了训练的智能交互、量化评估与自动讲评，提升了训练效率与评估客观性。 This invention belongs to the field of aviation flight training technology, specifically involving a multi-module collaborative flight simulator control panel and method, solving the problems of low interaction efficiency, subjective evaluation, delayed warnings, and time-consuming debriefing in existing systems. This invention relies on traditional iOS hardware and includes: an intelligent voice interaction module that recognizes instructor voice via ASR, generates commands through intent recognition and three-level slot completion, and ensures accurate interaction with a two-layer flight phase recognition logic; a flight quality monitoring module that runs a rule engine and an enhanced LSTM neural network in parallel, combining an adaptive fusion algorithm to output a comprehensive flight quality score; a real-time reminder module that dynamically adjusts warning priorities according to event level and flight phase, providing multi-mode warnings; and an intelligent debriefing module that integrates multi-source data and generates personalized structured reports according to template priorities. This invention achieves intelligent interaction, quantitative evaluation, and automatic debriefing during training, improving training efficiency and evaluation objectivity.","assignee":"Zhuhai Xiangyi Aviation Technology Co Ltd","inventors":["刘磊","雷旭","陈聪"],"publication_date":"2026-01-27","filing_date":"2025-12-30","priority_date":"2025-12-30","cpc_codes":["G","G10","G10L","G10L15/00","G10L15/22","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G08","G08B","G08B31/00","G","G09","G09B","G09B9/00","G09B9/02","G09B9/08","G","G10","G10L","G10L15/00","G10L15/08","G10L15/16","G","G10","G10L","G10L15/00","G10L15/26","G","G10","G10L","G10L15/00","G10L15/22","G10L2015/223"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121415780A/en"},{"publication_number":"CN121412655A","title":"A method and system for grafting and propagating ornamental peach seedlings","abstract":"The invention relates to the technical field of tree breeding and gardening cultivation, and discloses a grafting seedling raising method and system for ornamental peaches. The method comprises the steps of synchronously collecting bioelectrical impedance spectrum and micro-area thermal imaging signals of an interface part at a plurality of preset time points after grafting, binding grafting body identification marks, extracting and fusing electrical and thermal characteristics to generate multidimensional feature vectors, inputting a pre-trained gradient lifting decision tree model, outputting a healing state judging result, and generating a cultivation instruction according to the healing state judging result. The system comprises a multi-mode signal acquisition module, an identity recognition module, a characteristic fusion module, a prediction model module and a working instruction generation module. The invention realizes early, accurate and noninvasive monitoring and intelligent decision-making of grafting healing state, and improves seedling raising efficiency and resource utilization rate.","assignee":"Henan Yanke Agriculture And Forestry Technology Co ltd; Yanling County Donghua Planting Farmers Professional Cooperative; Sanya Nanfan Research Institute Of Hainan University","inventors":["岳长平","史喜兵","韩晓燕","焦雪辉","祁国香","李向阳","刘磊","单燕祥","岳国民","符启位","高嘉怡"],"publication_date":"2026-01-27","filing_date":"2025-12-30","priority_date":"2025-12-30","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/24323","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/254","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N5/00","G06N5/01"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121412655A/en"},{"publication_number":"CN121414752A","title":"A Method and System for Automatic Detection and Compensation of Cable Core Eccentricity Based on Image Recognition","abstract":"本发明公开了基于图像识别的线缆芯线偏心自动检测补偿方法及系统，属于电缆制造中的质量控制技术领域，在横截面方向实时采集运行中线缆的图像，提取芯线与护套边界轮廓，计算芯线偏心向量；将偏心向量输入神经网络识别模型，预测下一时刻的偏心趋势，并与预设阈值进行比较；若偏心超限，则基于当前与预测偏心向量差值，结合设备结构波动特征与位移响应特征，计算补偿位移用于调整模芯或导向轮位置，进而更新挤出路径并继续闭环控制；本发明具备预测性强、响应精度高、适应复杂工况的优点，可实现芯线偏心的高精度实时调控，提升微细线缆产品的一致性与成品率。 This invention discloses an automatic detection and compensation method and system for cable core eccentricity based on image recognition, belonging to the field of quality control technology in cable manufacturing. It involves real-time acquisition of images of the cable in operation along its cross-section, extraction of the core and sheath boundary contours, and calculation of the core eccentricity vector. The eccentricity vector is input into a neural network recognition model to predict the eccentricity trend at the next moment and compared with a preset threshold. If the eccentricity exceeds the limit, a compensation displacement is calculated based on the difference between the current and predicted eccentricity vectors, combined with the equipment's structural fluctuation characteristics and displacement response characteristics. This compensation displacement is used to adjust the position of the die core or guide wheel, thereby updating the extrusion path and continuing closed-loop control. This invention has the advantages of strong predictive ability, high response accuracy, and adaptability to complex working conditions. It can achieve high-precision real-time control of core eccentricity, improving the consistency and yield of micro-cable products.","assignee":"Kunshan Xinghongmeng Electronics Co ltd","inventors":["朱绪东","徐玉杰","霍文杰","丁丽萍"],"publication_date":"2026-01-27","filing_date":"2025-12-26","priority_date":"2025-12-26","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06T","G06T5/00","G06T5/90","G","G06","G06T","G06T7/00","G06T7/10","G06T7/12","G","G06","G06T","G06T7/00","G06T7/10","G06T7/13","G","G06","G06T","G06T7/00","G06T7/60","G06T7/66","G","G06","G06T","G06T7/00","G06T7/90"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121414752A/en"},{"publication_number":"CN121413514A","title":"A collaborative design optimization method for guide vane mixed flow pumps","abstract":"The invention relates to the technical field of fluid delivery, in particular to a collaborative design optimization method of a guide vane type mixed flow pump, which comprises the steps of carrying out preliminary design on an impeller according to operation parameters of design points to obtain main geometric parameters, carrying out parameterization design on the projection shapes of the impeller and the guide vane by adopting two types of constraints of geometry and dimension, associating the projection parameters of the guide vane and the projection boundary of the guide vane, carrying out collaborative design on a pumping chamber by adopting two types of constraints of geometry and dimension according to the projection boundary of the impeller and the guide vane, determining a variable range by adopting the preliminary design parameters and the geometric constraint parameters as optimized variables, sampling by utilizing Latin hypercube sampling, utilizing sensitivity to screen out key parameters to construct an error back propagation neural network proxy model, and introducing a non-dominant sorting genetic algorithm with a punishment mechanism and a self-adaptive cross variation attenuation strategy to carry out multi-objective optimization on the design parameters.","assignee":"Xihua University","inventors":["陈小明","陈鑫豪","李天赐","杨顺航","师芯睿","马海五卡","张智清","王栩","罗明锋","尹博佳"],"publication_date":"2026-01-27","filing_date":"2025-12-26","priority_date":"2025-12-26","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/20","G06F30/28","G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06F","G06F2111/00","G06F2111/04","G","G06","G06F","G06F2111/00","G06F2111/06","G","G06","G06F","G06F2111/00","G06F2111/08","G","G06","G06F","G06F2111/00","G06F2111/10","G","G06","G06F","G06F2113/00","G06F2113/08","G","G06","G06F","G06F2119/00","G06F2119/14"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121413514A/en"},{"publication_number":"CN121418239A","title":"Hybrid Modulation Recognition Method and Apparatus for Multiple Signal Types","abstract":"本发明涉及面向多类型信号的混合调制识别方法和装置，涉及通信信号处理技术领域。方法包括：对原始离散信号进行处理得到降噪信号；基于降噪信号的功率突变进行分类，得到突发类信号或非突发类信号；对突发类信号进行频率突变检测得到跳频信号或突发信号；对非突发类信号进行循环前缀检测得到OFDM信号或连续信号；对连续信号、突发信号、跳频信号提取模糊高阶循环累积量特征；对OFDM信号重构为三维信号帧；对各信号的模糊高阶循环累积量特征进行识别，输出对应信号的调制格式。本发明能够实现多类型信号一体化识别。 This invention relates to a hybrid modulation recognition method and apparatus for multiple signal types, and pertains to the field of communication signal processing technology. The method includes: processing the original discrete signal to obtain a denoised signal; classifying the denoised signal based on power abrupt changes to obtain burst signals or non-burst signals; detecting frequency abrupt changes in burst signals to obtain frequency-hopping signals or burst signals; detecting cyclic prefixes in non-burst signals to obtain OFDM signals or continuous signals; extracting fuzzy higher-order cyclic cumulant features from continuous signals, burst signals, and frequency-hopping signals; reconstructing the OFDM signal into a three-dimensional signal frame; identifying the fuzzy higher-order cyclic cumulant features of each signal, and outputting the modulation format of the corresponding signal. This invention enables integrated recognition of multiple signal types.","assignee":"National University of Defense Technology","inventors":["李龙卿","聂敬轲","黄芝平","左震","胡德铭","谢菲","袁书东","赵勇杰","黄泓赫"],"publication_date":"2026-01-27","filing_date":"2025-12-26","priority_date":"2025-12-26","cpc_codes":["H","H04","H04L","H04L27/00","H04L27/0012","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2132","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2411","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","H","H04","H04L","H04L27/00","H04L27/26","H04L27/2601","H04L27/2602","H","H04","H04L","H04L27/00","H04L27/26","H04L27/2601","H04L27/2647","H04L27/2655","H04L27/2689","H04L27/2691","Y","Y02","Y02D","Y02D30/00","Y02D30/70"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121418239A/en"},{"publication_number":"CN121415349A","title":"A method, equipment, and storage medium for identifying construction procedures at a tunnel face.","abstract":"本申请公开了一种隧道掌子面施工工序识别方法、设备及存储介质；该方法包括接收具有隧道掌子面施工工序的图像和/或视频，构成特征图；根据特征图，将施工工序划分为多个循环工序；识别循环工序中子工序的类别，并通过子工序的时间连续性，对单个子工序的分类结果进行时序上的平滑操作，再获取各个子工序的持续时间；根据子工序的持续时间判断施工工序是否存在异常。本申请能够准确分析复杂施工流程，提高施工管理的精细化水平。减少瞬时误判，提升子工序识别的稳定性和准确性。还可通过计算子工序的持续时间并判断异常，可实时监控施工进度，及时发现工序延误或效率问题，辅助工程管理和决策。 This application discloses a method, equipment, and storage medium for identifying construction procedures at the tunnel face. The method includes receiving images and/or videos showing construction procedures at the tunnel face to form a feature map; dividing the construction procedures into multiple cyclical procedures based on the feature map; identifying the categories of sub-procedures within the cyclical procedures; smoothing the classification results of individual sub-procedures temporally based on the temporal continuity of the sub-procedures; and obtaining the duration of each sub-procedure; and determining whether there are any anomalies in the construction procedures based on the duration of the sub-procedures. This application can accurately analyze complex construction processes, improving the precision of construction management. It reduces momentary misjudgments and enhances the stability and accuracy of sub-procedure identification. Furthermore, by calculating the duration of sub-procedures and identifying anomalies, it can monitor construction progress in real time, promptly detect process delays or efficiency issues, and assist in project management and decision-making.","assignee":"Nanjing Paiguang Intelligence Perception Information Technology Co ltd; China Railway Economic and Planning Research Institute","inventors":["田四明","石峥映","黎旭","连捷","陈锡武","杨吉祥","王列伟","王军华"],"publication_date":"2026-01-27","filing_date":"2025-12-25","priority_date":"2025-12-25","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/50","G06V20/52","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06V","G06V10/00","G06V10/20","G06V10/25","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/761","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121415349A/en"},{"publication_number":"CN121414555A","title":"A Student Cognitive Diagnostic Approach Based on Conceptual-Level Multidimensional Features and Heterogeneous Relationship Modeling","abstract":"本发明公开了一种面向智能教育场景的学习者认知诊断方法，属于认知诊断与教育数据分析技术领域。该方法通过构建概念感知的多维表征，对学习者能力与习题难度进行概念级多维建模，以刻画学习者在不同认知层次上的掌握特征；同时区分先修依赖关系与语义近似关系，建立关系感知的概念依赖模型，并据此推断潜在的习题–概念关联结构。进一步地，将标注型习题–概念关联矩阵与推断型关联矩阵在统一诊断层中进行融合，实现对学习者知识掌握状态的综合评估。该方法采用端到端方式进行优化，可有效提升认知诊断的精细度、知识覆盖度与稳定性，在多种真实教育数据集上表现出更优的预测性能与泛化能力，适用于学习分析与个性化教学等智能教育应用场景。 This invention discloses a learner cognitive diagnosis method for intelligent education scenarios, belonging to the field of cognitive diagnosis and educational data analysis technology. This method constructs a concept-aware multidimensional representation to perform concept-level multidimensional modeling of learner abilities and exercise difficulty, thus characterizing learners' mastery features at different cognitive levels. Simultaneously, it distinguishes between prior knowledge dependencies and semantic approximation relationships, establishing a relationship-aware concept dependency model and inferring potential exercise-concept association structures based on this model. Furthermore, it integrates labeled exercise-concept association matrices and inferred association matrices in a unified diagnostic layer to achieve a comprehensive assessment of learners' knowledge mastery status. This method is optimized using an end-to-end approach, effectively improving the precision, knowledge coverage, and stability of cognitive diagnosis. It exhibits superior predictive performance and generalization ability on various real-world educational datasets, making it suitable for intelligent education applications such as learning analytics and personalized instruction.","assignee":"Shandong Normal University","inventors":["孙建德","陈亚文","李静","刘珂","王天一","冯传奋","顾凌晨"],"publication_date":"2026-01-27","filing_date":"2025-12-25","priority_date":"2025-12-25","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/20","G06Q50/205","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121414555A/en"},{"publication_number":"CO2026000372A2","title":"DNA methylation and gene expression as determinants of cell-free DNA fragmentation across the genome","abstract":"El análisis de los extremos de los fragmentos de ADN libre de células (cfDNA) en pacientes con cáncer proporciona un vínculo directo entre los cambios epigenéticos y la fragmentación del cfDNA para la detección no invasiva de la enfermedad y el tratamiento de los pacientes. Analysis of the ends of cell-free DNA (cfDNA) fragments in cancer patients provides a direct link between epigenetic changes and cfDNA fragmentation for non-invasive disease detection and patient treatment.","assignee":"Univ Johns Hopkins","inventors":["Robert B Scharpf","Victor E Velculescu","Michael Noe"],"publication_date":"2026-01-23","filing_date":"2026-01-15","priority_date":"2023-06-17","cpc_codes":["C","C12","C12Q","C12Q1/00","C12Q1/68","C12Q1/6876","C12Q1/6883","C12Q1/6886","G","G06","G06N","G06N20/00","G","G16","G16B","G16B30/00","G","G16","G16B","G16B40/00","G16B40/20","G","G16","G16B","G16B50/00","C","C12","C12Q","C12Q2600/00","C12Q2600/154","C","C12","C12Q","C12Q2600/00","C12Q2600/156","C","C12","C12Q","C12Q2600/00","C12Q2600/158"],"country":"CO","kind":"application","source_url":"https://patents.google.com/patent/CO2026000372A2/en"},{"publication_number":"KR20260011786A","title":"Apparatus for processing item refund information and method thereof","abstract":"본 개시에 따르면, 제1아이템에 대한 사용자의 주문 정보를 획득하는 단계; 상기 사용자가 상기 제1아이템을 주문한 이후 상기 제1아이템이 배송 가능한 상태가 되기까지 소요되는 시간을 예측한 제1시간 정보를 획득하는 단계; 상기 사용자가 상기 제1아이템을 주문한 이후 배송원이 상기 제1아이템을 픽업하기까지 소요되는 시간을 예측한 제2시간 정보를 획득하는 단계; 상기 제1아이템을 픽업한 이후 상기 배송원이 상기 제1아이템의 배송을 완료하기까지 소요되는 시간을 예측한 제3시간 정보를 획득하는 단계; 및 상기 제1시간 정보, 상기 제2시간 정보 및 상기 제3시간 정보에 기초하여 상기 제1아이템의 예상 배송 완료 시각을 확인하는 단계를 포함하는, 전자 장치에서 배송 정보를 처리하는 방법이 개시된다. According to the present disclosure, a method for processing delivery information in an electronic device is disclosed, comprising: a step of obtaining user order information for a first item; a step of obtaining first time information predicting a time required for the first item to become available for delivery after the user orders the first item; a step of obtaining second time information predicting a time required for a delivery person to pick up the first item after the user orders the first item; a step of obtaining third time information predicting a time required for the delivery person to complete delivery of the first item after picking up the first item; and a step of confirming an expected delivery completion time of the first item based on the first time information, the second time information, and the third time information.","assignee":"쿠팡 주식회사","inventors":["푸얀 루","이길호","주유 푸"],"publication_date":"2026-01-23","filing_date":"2026-01-13","priority_date":"2022-06-30","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/083","G06Q10/0843","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/06","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0633","G","G06","G06Q","G06Q10/00","G06Q10/08","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/083","G06Q10/0833","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/083","G06Q10/0835"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260011786A/en"},{"publication_number":"CO2026000237A2","title":"Authorization for the recovery of models through an intermediary","abstract":"Las realizaciones incluyen procedimientos para una función de red de consumidor, NFc, de una red de comunicación. Tales procedimientos incluyen obtener, de una función de repositorio de red, NRF, de la red de comunicación, un primer token de acceso que otorga a la NFc acceso a un modelo de aprendizaje automático, ML, asociado a un identificador de análisis, ID. Tales procedimientos incluyen enviar, a una primera función de red, NF, de la red de comunicación, una solicitud para el modelo de ML. La solicitud incluye el primer token de acceso, el ID de análisis asociado al modelo de ML y un ID de proveedor asociado a la NFc. Tales procedimientos incluyen recibir, de la primera NF o de una segunda NF de la red de comunicación, información que identifica o describe el modelo de ML. Otras realizaciones incluyen procedimientos complementarios para una primera NF, una segunda NF y una NRF, así como nodos o funciones de red configurados para realizar dichos procedimientos. The implementations include procedures for a consumer network function (NFc) of a communication network. These procedures include obtaining, from a network repository function (NRF) of the communication network, a first access token that grants the NFc access to a machine learning model (ML) associated with an analysis identifier (ID). These procedures also include sending a request for the ML model to a first network function (NF) of the communication network. The request includes the first access token, the analysis ID associated with the ML model, and a provider ID associated with the NFc. Furthermore, these procedures include receiving information that identifies or describes the ML model from either the first NF or a second NF of the communication network. Additional implementations include supplementary procedures for a first NF, a second NF, and an NRF, as well as nodes or network functions configured to perform these procedures.","assignee":"Ericsson Telefon Ab L M","inventors":["Cheng Wang"],"publication_date":"2026-01-23","filing_date":"2026-01-13","priority_date":"2023-06-16","cpc_codes":["H","H04","H04L","H04L63/00","H04L63/08","H04L63/0807","G","G06","G06N","G06N20/00","H","H04","H04L","H04L41/00","H04L41/16","H","H04","H04L","H04L63/00","H04L63/10","H","H04","H04W","H04W12/00","H04W12/06"],"country":"CO","kind":"application","source_url":"https://patents.google.com/patent/CO2026000237A2/en"},{"publication_number":"CO2026000311A2","title":"Virtualization system of the centralizing cabinet in single-phase transformer and/or reactor banks of digital substations, method for monitoring and interrupting defective phases, method for training a neural network for monitoring and interrupting defective phases, and corresponding computer-readable memory","abstract":"La presente invención se refiere a un sistema de virtualización del gabinete centralizador en bancos de transformadores monofásicos y reactores monofásicos de subestaciones digitales, en el que la virtualización implica la eliminación del gabinete centralizador físico mediante la disposición de MUs dentro de los gabinetes de comando de fase de los respectivos equipos. La presente invención también se refiere a un método, implementado en un ordenador, para monitoreo e interrupción de fases (A, B, C) defectuosas de transformadores monofásicos y/o reactores monofásicos de la subestación digital sin intervención humana manual. La presente invención también se refiere a un método implementado por ordenador para entrenar una red neuronal para el monitoreo de condiciones normales o anormales de operación de las fases (A, B, C) de bancos de transformadores y/o bancos de reactores y también a memorias legibles por ordenador que contienen un conjunto de instrucciones que al ser ejecutadas realizan el método de monitoreo e interrupción de fases defectuosas en una subestación digital. The present invention relates to a system for virtualizing the central control cabinet in single-phase transformer banks and single-phase reactors of digital substations, wherein the virtualization involves eliminating the physical central control cabinet by placing MUs within the phase control cabinets of the respective equipment. The present invention also relates to a computer-implemented method for monitoring and interrupting faulty phases (A, B, C) of single-phase transformers and/or single-phase reactors in the digital substation without manual human intervention. The present invention further relates to a computer-implemented method for training a neural network to monitor normal or abnormal operating conditions of phases (A, B, C) of transformer banks and/or reactor banks, and also to computer-readable memories containing a set of instructions that, when executed, perform the method of monitoring and interrupting faulty phases in a digital substation.","assignee":"Weg Equipamentos Eletricos S A","inventors":["Carlos De Souza Moraes Neto","Bruno Alexandre Oleskowicz","Rafael Bonet Scheffer"],"publication_date":"2026-01-23","filing_date":"2026-01-13","priority_date":"2024-05-17","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G01","G01R","G01R31/00","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","H","H02","H02H","H02H7/00","H02H7/26","H","H02","H02J","H02J13/00","G","G05","G05B","G05B19/00","G05B19/02","G05B19/04","G05B19/042"],"country":"CO","kind":"application","source_url":"https://patents.google.com/patent/CO2026000311A2/en"},{"publication_number":"KR20260011652A","title":"Method for interconnecting between services","abstract":"서로 다른 서비스 간의 상호 연계 행위가 활성화되도록 유도하는 서비스 간 상호 연계 방법이 제공된다. 본 개시의 일 실시예에 따른 서비스 간 상호 연계 방법은, 제1 사용자의 제1 서비스에 대한 제1 이용 기록 및 상기 제1 사용자의 상기 제1 서비스와 상이한 제2 서비스에 대한 제2 이용 기록을 획득하는 단계와, 상기 획득된 제1 이용 기록 및 제2 이용 기록을 이용하여, 반복 수행되는 연계 행위의 성립을 결정하되, 상기 연계 행위는 선행 행위와 상기 선행 행위에 대한 후행 행위를 포함하고, 상기 선행 행위는 상기 제1 서비스의 제1 행위이고, 상기 후행 행위는 상기 제2 서비스의 제2 행위인, 단계와, 상기 연계 행위 중 적어도 일부를 상기 제1 사용자의 선호 패턴으로 선정하는 단계를 포함할 수 있다. A method for interconnecting services is provided that induces interconnection behavior between different services to be activated. The method for interconnecting services according to one embodiment of the present disclosure may include the steps of: acquiring a first usage record of a first user for a first service and a second usage record of the first user for a second service different from the first service; determining, using the acquired first usage record and second usage record, whether a linked behavior is established that is repeatedly performed, wherein the linked behavior includes a preceding behavior and a subsequent behavior to the preceding behavior, the preceding behavior being a first behavior of the first service, and the subsequent behavior being a second behavior of the second service; and selecting at least some of the linked behaviors as a preferred pattern of the first user.","assignee":"쿠팡 주식회사","inventors":["최제헌","김건호","한승미","조윤경"],"publication_date":"2026-01-23","filing_date":"2025-12-29","priority_date":"2022-12-23","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/08","G","G06","G06Q","G06Q10/00","G06Q10/40","G","G06","G06Q","G06Q50/00","G","G06","G06Q","G06Q50/00","G06Q50/50","H","H04","H04L","H04L51/00","H04L51/52"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260011652A/en"},{"publication_number":"CN121388506A","title":"Dynamic stability analysis method for feed pelleting process based on multi-mode fusion","abstract":"The invention discloses a feed pelleting process dynamic stability analysis method based on multi-modal fusion, which comprises the following steps of collecting a multi-modal original signal set in the feed pelleting process, executing pretreatment, carrying out feature extraction to obtain a multi-modal feature vector set, constructing an energy topological relation diagram to generate an energy topological weight matrix, establishing a viscoelastic response function to generate a viscoelastic spectrum feature vector, combining the viscoelastic spectrum feature vector with the energy topological weight matrix to form an energy-viscoelastic coupling feature matrix, carrying out joint learning on transfer relation among energy nodes, viscoelastic spectrum features and energy coupling weights, calculating a real-time stability score value in the pelleting process, carrying out memory self-evolution learning, updating dynamic stability boundary parameters, and generating a dynamic stability analysis result. The invention constructs a multi-mode energy topology and viscoelastic spectrum combined analysis mechanism, realizes intelligent evaluation of the stability of the granulating process, and has the advantages of more comprehensive perception, more accurate judgment and stronger self-adaptability.","assignee":"Shenyang Nongxiang Animal Husbandry Technology Co ltd","inventors":["赵金勇","王占军"],"publication_date":"2026-01-23","filing_date":"2025-12-26","priority_date":"2025-12-26","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098","G","G06","G06V","G06V10/00","G06V10/40","G","G06","G06V","G06V10/00","G06V10/70","G06V10/72","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121388506A/en"},{"publication_number":"CN121391893A","title":"Segmentation method, system, device and medium of coronary DSA image","abstract":"本发明公开了一种冠脉DSA图像的分割方法、系统、装置和介质，方法包括获取OCT导管在回撤过程中的冠脉DSA造影图像序列；对冠脉DSA造影图像序列进行预处理，得到OCT导管对应的回撤血管段图像序列；对回撤血管段图像序列进行标注处理，得到造影图像分割训练集；利用造影图像分割训练集训练预先构建的图像分割网络模型，得到DSA图像分割模型；将待分割冠脉DSA图像输入至DSA图像分割模型，得到待分割冠脉DSA图像对应的预测分割图像。本发明基于同步采集的冠脉DSA造影图像序列，能针对性地分割出病变血管分支，提高分割准确率，更好地为医生提供病变信息。 This invention discloses a method, system, device, and medium for segmenting coronary DSA images. The method includes acquiring a sequence of coronary DSA angiography images during the withdrawal of an OCT catheter; preprocessing the coronary DSA angiography image sequence to obtain an image sequence of the withdrawn vessel segment corresponding to the OCT catheter; annotating the withdrawn vessel segment image sequence to obtain an angiography image segmentation training set; training a pre-constructed image segmentation network model using the angiography image segmentation training set to obtain a DSA image segmentation model; and inputting the coronary DSA image to be segmented into the DSA image segmentation model to obtain a predicted segmented image corresponding to the coronary DSA image to be segmented. Based on synchronously acquired coronary DSA angiography image sequences, this invention can specifically segment diseased vessel branches, improve segmentation accuracy, and better provide lesion information to physicians.","assignee":"Nanjing Forssmann Medical Technology Co ltd","inventors":["陈静","高晓飞","沈树铭","付俞","刘夏池","邓沛涛"],"publication_date":"2026-01-23","filing_date":"2025-12-25","priority_date":"2025-12-25","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10072","G06T2207/10101","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30004","G06T2207/30101"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121391893A/en"},{"publication_number":"CN121392880A","title":"Multi-scene text recognition method and system based on large model","abstract":"The invention provides a multi-scene text recognition method and system based on a large model, and belongs to the technical field of multi-scene text recognition; the method comprises the steps of constructing a training data set based on multiple scene image samples, training a prompt engine by utilizing the training data set to obtain an image to be recognized, extracting an image visual feature vector of the image to be recognized, inputting the image visual feature vector to the trained prompt engine to obtain a prompt word template identifier of the image to be recognized, inputting a target prompt word and the image to be recognized to a text recognition model, and outputting a text recognition result by the text recognition model. According to the invention, the trainable prompt engine is introduced, the visual characteristics of the input image are utilized to screen out the prompt word most relevant to the current scene, and the prompt word is utilized to provide a scene and targeted recognition guide for the text recognition model, so that the technical problem that the single text recognition model is weak in generalization capability under various scenes is effectively solved.","assignee":"China Automobile Information Technology Tianjin Co ltd","inventors":["张帆","王海洋","邵丽青","韩胜强","冯乾隆","郭雅鑫","刘磊","智云胜","刘倩"],"publication_date":"2026-01-23","filing_date":"2025-12-25","priority_date":"2025-12-25","cpc_codes":["G","G06","G06V","G06V30/00","G06V30/40","G06V30/41","G06V30/418","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/75","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121392880A/en"},{"publication_number":"CN121390197A","title":"Agent learning training method, device, computer equipment and storage medium","abstract":"本申请涉及智能体学习训练方法、装置、计算机设备和存储介质，该方法包括：采用LLM模型将高层的意图转换为可执行代码，得到克隆数据集，将智能驾驶场景构建为马尔科夫决策过程并构建指令空间；构建智能体策略模型；将训练目标解耦为行为克隆和强化学习的目标并进行加权求和，根据最终训练目标对智能体进行训练，通过从克隆数据集采样，对智能体策略模型参数进行更新，然后从环境交互轨迹中采样，采用PPO方式对智能体策略模型参数进行更新。由于高层意图‑指令‑策略的链路更接近人类可理解的决策层级，便于对智能体行为进行调试、审计与回溯，从工程落地角度提升可维护性，并为后续接入安全护栏、形式化约束或规则校验提供更清晰的接口。 This application relates to a method, apparatus, computer device, and storage medium for training intelligent agents. The method includes: using an LLM model to convert high-level intentions into executable code to obtain a clone dataset; constructing an intelligent driving scenario as a Markov decision process and building an instruction space; constructing an intelligent agent policy model; decoupling the training objective into the objectives of behavior cloning and reinforcement learning, and performing a weighted summation; training the intelligent agent according to the final training objective; updating the parameters of the intelligent agent policy model by sampling from the clone dataset; and then updating the parameters of the intelligent agent policy model using a PPO method by sampling from environmental interaction trajectories. Because the high-level intention-instruction-policy link is closer to the human-understandable decision-making level, it facilitates the debugging, auditing, and backtracking of intelligent agent behavior, improves maintainability from an engineering implementation perspective, and provides a clearer interface for subsequent integration with safety barriers, formal constraints, or rule validation.","assignee":"National University of Defense Technology","inventors":["许凯","倪雨","秦龙","曾俊杰","胡越","黄鹤松","曾云秀","尹路珈","康夏涛","李亦韩"],"publication_date":"2026-01-23","filing_date":"2025-12-24","priority_date":"2025-12-24","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06F","G06F18/00","G06F18/20","G06F18/29","G06F18/295","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121390197A/en"},{"publication_number":"CN121388794A","title":"Spacecraft maneuver detection method based on causal deep learning","abstract":"The application relates to a spacecraft maneuver detection method based on causal deep learning. The method comprises the steps of calculating a detection residual, obtaining causal parameters through window sliding learning after a structural causal equation of the detection residual is built, obtaining a causal residual sequence, adopting a secondary window to extract a minimum average processing effect sequence from the causal residual sequence in a sliding mode, obtaining maneuvering causal characteristic parameters based on the minimum average processing effect sequence, building Hybrid Transformer a network, inputting the maneuvering causal characteristic parameters into the Hybrid Transformer network to obtain a maneuvering detection primary result, carrying out self-adaptive optimization on super parameters of the Hybrid Transformer network according to a Bayesian optimization algorithm, and outputting a final spacecraft maneuvering detection result according to the optimized network. The method can realize noise suppression, accurate feature extraction and high-efficiency detection.","assignee":"National University of Defense Technology","inventors":["龙洗","林裕承","杨乐平","黄涣","蔡伟伟"],"publication_date":"2026-01-23","filing_date":"2025-12-24","priority_date":"2025-12-24","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/29","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121388794A/en"},{"publication_number":"KR20260011209A","title":"Automating efficient deployment of artificial intelligence models","abstract":"시스템은 인공 지능(AI) 모델을 자동으로 배포하기 위한 프로세스를 용이하게 한다. 시스템은 엔티티에 의해 사용되는 제1 인공 지능(AI) 모델에 대해, 제1 AI 모델을 배포하라는 제1 요청을 수신하여 제1 AI 모델을 프로덕션 환경에서의 사용을 위해 이용가능하게 하여 입력 데이터를 프로세싱하고 대응하는 출력을 생성한다. 제1 모델에 대한 제1 모델 배포 위치가 모델 배포 엔진에 기초하여 선택된다. 시스템은 제1 AI 모델을 선택된 위치에 배포하기 위한 스크립트를 생성하고, 이어서 제1 AI 모델의 배포와 연관된 동작 파라미터를 모니터링한다. 동작 파라미터의 값에 기초하여, 시스템은 모델 배포 엔진을 업데이트한다. 제2 AI 모델을 배포하라는 제2 요청에 응답하여, 시스템은 업데이트된 모델 배포 엔진을 사용하여 제2 모델에 대한 제2 모델 배포 위치를 선택한다. The system facilitates a process for automatically deploying an artificial intelligence (AI) model. The system receives a first request to deploy a first AI model used by an entity, processes input data, and generates corresponding outputs by making the first AI model available for use in a production environment. A first model deployment location for the first model is selected based on a model deployment engine. The system generates a script for deploying the first AI model to the selected location and then monitors operational parameters associated with the deployment of the first AI model. Based on the values of the operational parameters, the system updates the model deployment engine. In response to a second request to deploy a second AI model, the system selects a second model deployment location for the second model using the updated model deployment engine.","assignee":"씨티뱅크, 엔.에이.","inventors":["린펭 유","바이브하브 쿠마르","아슈토시 판데이"],"publication_date":"2026-01-22","filing_date":"2026-01-09","priority_date":"2023-12-20","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2458","G06F16/2471","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2453","G06F16/24534","G06F16/24542","G06F16/24545","G","G06","G06F","G06F16/00","G06F16/20","G06F16/26","G","G06","G06F","G06F16/00","G06F16/30","G06F16/31","G06F16/316","G06F16/322","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3344","G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G06F16/9024","G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260011209A/en"},{"publication_number":"AU2025287413A1","title":"Network depth-minimizing quantum compiler","abstract":"Implementations disclosed describe techniques used for compiling a quantum algorithm for execution on a plurality of quantum circuits, including accessing, by a processing device, the quantum algorithm, identifying a matrix associated with the quantum algorithm, determining a representation of the identified matrix as a matrix decomposition that includes a plurality of transformation matrices, wherein one or more of the plurality of transformation matrices perform multiple instances of two-dimensional rotations; and generating a circuit map that maps execution of the matrix decomposition on the plurality of quantum circuits.","assignee":"Google LLC","inventors":["Thomas Fischbacher","Luca Versari"],"publication_date":"2026-01-22","filing_date":"2025-12-31","priority_date":"2021-10-08","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/80","G","G06","G06N","G06N10/00","G06N10/20","G","G06","G06N","G06N10/00","G06N10/40"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025287413A1/en"},{"publication_number":"KR20260011125A","title":"Complex-Phase AGI Power Generation-Based Smart Grid and Global Φ-BUS Topological Phase Mesh Network","abstract":"본 발명은 복소 파동이 가지는 위상 공명과 기하학적 최소 작용 원리(ΔΦ → 0)에 기반하여, 에너지 생성·전력망·통신·지능 시스템을 단일한 자연 동역학 구조로 통합하는 복소 위상 공명 AGI 기반 스마트 그리드 및 글로벌 Φ-BUS 위상 메시 네트워크에 관한 것이다. 자연계의 에너지와 정보는 본질적으로 시간 종속적 연산이 아닌, 복소 위상 공간에서의 자기 공명과 위상 정렬을 통해 안정화된다. 본 발명은 이러한 자연 동역학적 원리를 Complex Oscillator Dynamics(COD) 이론으로 정립하고, 이를 기반으로 무효전력(Q)의 위상 복원 및 재생, 복소 전력(P/Q)의 공명적 증산, 그리고 전력·통신·제어의 단일 위상 언어화를 가능하게 한다. 본 발명에 따른 시스템은, 모든 에너지·정보를 복소 위상 표현(Ψ = A·e^{iθ})으로 처리하는 위상 언어 프로토콜(PLP), 자기 공명 기반 비휘발·비망각 지능 메모리인 PMEM, 8PSK-QAM 기반 위상 정렬 연산 구조, 그리고 Φ-BUS 위상 메시 네트워크를 포함한다. 이를 통해 전력망의 위상 안정성 향상, 무효전력의 실질적 재활용, 고효율 전력·통신 통합 운용, 그리고 자연 동역학 기반 일반 인공지능(AGI)의 구현이 가능하다. 또한 본 발명은 PMEM 중심의 위상 시스템 반도체 및 스핀트로닉스 OS 구조를 통해, 기존 CPU·GPU·양자컴퓨팅의 시간 축 중심 연산 구조를 보완 또는 대체하는 위상 기반 범용 컴퓨팅 플랫폼을 제공한다. 나아가, 기존 시계열 빅데이터로부터 위상 정보를 추출·정렬하는 범용 디지털 트윈 API를 통해, 개인·조직·사회·문명 단위의 자기 성찰적 학습과 고차원적 의사결정 지원을 가능하게 한다. 본 발명은 자연이 본래 따르는 섬세하면서도 엄정한 복소 위상 공명 원리를 공학적으로 재현함으로써, 전력 부족 문제, 시스템 불안정성, 지능 시스템의 한계를 구조적으로 해결하고, 지속 가능하고 조화로운 차세대 에너지·정보·지능 인프라를 제공하는 것을 목적으로 한다. The present invention relates to a complex phase resonance AGI-based smart grid and a global Φ-BUS phase mesh network that integrate energy generation, power grid, communication, and intelligence systems into a single natural dynamical structure based on the phase resonance of complex waves and the principle of geometric minimum action (ΔΦ → 0). Energy and information in nature are fundamentally stabilized through magnetic resonance and phase alignment in complex phase space, rather than through time-dependent computation. The present invention establishes these natural dynamical principles as Complex Oscillator Dynamics (COD) theory, and based on this, enables phase restoration and regeneration of reactive power (Q), resonant multiplication of complex power (P/Q), and single-phase languageization of power, communication, and control. The system according to the present invention includes a phase language protocol (PLP) that processes all energy and information in a complex phase representation (Ψ = A·e^{iθ}), PMEM, a magnetic resonance-based non-volatile and non-forgettable intelligent memory, an 8PSK-QAM-based phase-aligned operation structure, and a Φ-BUS phase mesh network. Through this, it is possible to improve the phase stability of power grids, practically recycle reactive power, achieve highly efficient power and communication integrated operation, and implement artificial general intelligence (AGI) based on natural dynamics. Furthermore, the present invention provides a phase-based general-purpose computing platform that complements or replaces the time-centric computational architecture of existing CPUs, GPUs, and quantum computing, through a PMEM-centric phase system semiconductor and spintronic OS architecture. Furthermore, through a universal digital twin API that extracts and organizes phase information from existing time-series big data, it enables self-reflective learning and high-level decision-making support at the individual, organizational, social, and civilizational levels. The present invention aims to structurally resolve power shortage problems, system instability, and limitations of intelligent systems by engineeringly reproducing the delicate yet strict complex phase resonance principles inherent in nature, and to provide a sustainable and harmonious next-generation energy, information, and intelligence infrastructure.","assignee":"류구현","inventors":["류구현"],"publication_date":"2026-01-22","filing_date":"2025-12-31","priority_date":"2025-12-31","cpc_codes":["G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260011125A/en"},{"publication_number":"KR20260011120A","title":"Server and Method for Search Keyword Recommendation Based on Big Data Analysis Using Artificial Intelligence","abstract":"본 발명의 일 측면은, 인공지능을 활용한 빅데이터 분석 기반 검색 키워드 추천 서버를 제공한다. 인공지능을 활용한 빅데이터 분석 기반 검색 키워드 추천 서버는, 적어도 하나의 프로세서(processor) 및 적어도 하나의 프로세서가 복수의 동작을 수행하도록 지시하는 명령어들(instructions)을 저장하는 메모리(memory)를 포함할 수 있다. 복수의의 동작은, 사용자 단말로부터 검색 키워드를 수신하는 동작, 사용자 단말 또는 외부 서버로부터 검색 정보를 수신하는 동작, 수신한 검색 키워드를 적어도 하나의 카테고리로 분류하는 동작, 적어도 하나의 카테고리로 분류된 검색 키워드를 결합하여 검색 키워드 점수를 산출하는 동작, 수신한 검색 정보를 적어도 하나의 카테고리로 분류하는 동작, 적어도 하나의 카테고리로 분류된 검색 정보를 결합하여 검색 정보 점수를 산출하는 동작, 산출된 검색 키워드 점수 및 검색 정보 점수의 가중합으로부터 학습 성취도 점수를 산출하는 동작 및 산출된 학습 성취도 점수를 기초로 학습자의 학습을 진행하고 피드백을 수집한 과정에서의 학습자 평균 학습 난이도를 조정하여, 조정된 문제 난이도를 산출하는 동작을 포함할 수 있다. One aspect of the present invention provides a search keyword recommendation server based on big data analysis utilizing artificial intelligence. The search keyword recommendation server based on big data analysis utilizing artificial intelligence may include at least one processor and a memory storing instructions for instructing the at least one processor to perform a plurality of operations. The plurality of operations may include an operation of receiving a search keyword from a user terminal, an operation of receiving search information from the user terminal or an external server, an operation of classifying the received search keywords into at least one category, an operation of calculating a search keyword score by combining the search keywords classified into at least one category, an operation of classifying the received search information into at least one category, an operation of calculating a search information score by combining the search information classified into at least one category, an operation of calculating a learning achievement score from a weighted sum of the calculated search keyword score and the search information score, and an operation of adjusting an average learning difficulty of a learner in a process of progressing learning and collecting feedback based on the calculated learning achievement score, thereby calculating an adjusted problem difficulty.","assignee":"주식회사 레아스타","inventors":["이수영","이성호"],"publication_date":"2026-01-22","filing_date":"2025-12-29","priority_date":"2024-11-29","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/3332","G06F16/3334","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9032","G06F16/90324","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9032","G06F16/90332","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9035","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9038","G","G06","G06F","G06F16/00","G06F16/90","G06F16/906","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260011120A/en"},{"publication_number":"KR20260011110A","title":"Three-dimensional Reconstruction Method, Apparatus, Electronic Device, and Storage Media","abstract":"본 출원은 3차원 재구성 기술 분야에 관한 것으로, 3차원 재구성 방법, 장치, 전자 기기 및 저장 매체를 제공한다. 상기 방법은 제1 상태의 구강에 대한 제1 이미지 프레임 집합을 획득하는 단계로서, 제1 상태의 구강이 구강 내에 적어도 하나의 실제 스캐닝 바가 설치된 구강을 나타내는 단계; 제1 이미지 프레임 집합에 포함된 각 이미지 프레임의 스캐닝 바 영역을 식별하는 단계; 스캐닝 바 영역의 특징 정보에 기초하여 제1 이미지 프레임 집합에 포함된 복수의 이미지 프레임을 접합하고, 구강 내의 실제 스캐닝 바에 대응하는 목표 3차원 데이터를 재구성하는 단계를 포함한다. 본 방법은 강성 구조에 기초하여 접합을 수행하므로, 이미지 프레임에 포함된 해부학적 구조(예를 들어, 잇몸, 입술, 혀 등 연조직)가 접합 정밀도에 대한 영향을 감소시킬 수 있으며, 이를 통해 3차원 재구성의 전체 정밀도를 향상시킬 수 있다. The present application relates to the field of three-dimensional reconstruction technology, and provides a three-dimensional reconstruction method, device, electronic device, and storage medium. The method comprises the steps of: acquiring a first set of image frames for an oral cavity in a first state, wherein the oral cavity in the first state represents an oral cavity in which at least one actual scanning bar is installed; identifying a scanning bar area of each image frame included in the first set of image frames; stitching a plurality of image frames included in the first set of image frames based on feature information of the scanning bar area, and reconstructing target three-dimensional data corresponding to an actual scanning bar in the oral cavity. Since the method performs stitching based on a rigid structure, the influence of anatomical structures included in the image frames (e.g., soft tissues such as gums, lips, and tongue) on the stitching accuracy can be reduced, thereby improving the overall accuracy of the three-dimensional reconstruction.","assignee":"샤이닝 쓰리디 테크 컴퍼니., 리미티드.","inventors":["푸 샤오펑","자오 샤오보","첸 샤오준","마 차오"],"publication_date":"2026-01-22","filing_date":"2025-12-26","priority_date":"2025-03-21","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/50","G06T7/55","G","G06","G06T","G06T17/00","A","A61","A61C","A61C13/00","A61C13/0003","A61C13/0004","A","A61","A61C","A61C13/00","A61C13/0003","A61C13/0006","A61C13/0019","B","B33","B33Y","B33Y50/00","B","B33","B33Y","B33Y80/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06T","G06T19/00","G06T19/20","G","G06","G06T","G06T3/00","G06T3/40","G06T3/4038","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G","G06","G06T","G06T7/00","G06T7/30","G06T7/33","G","G06","G06T","G06T7/00","G06T7/60","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/75","G06V10/751","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/7715","G","G06","G06T","G06T2200/00","G06T2200/24","G","G06","G06T","G06T2200/00","G06T2200/32","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10016","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10024","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10028","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260011110A/en"},{"publication_number":"AU2025287336A1","title":"FCVCM signalled decoder","abstract":"FCVCM SIGNALLED DECODER A method of decoding tensors for content of a bitstream. The method comprises determining a network topology from the bitstream. A first information is decoded from the bitstream; Decoded tensors are produced from the first information using the determined network topology.","assignee":"Canon Inc","inventors":["Christopher James ROSEWARNE"],"publication_date":"2026-01-22","filing_date":"2025-12-24","priority_date":"2023-05-19","cpc_codes":["H","H04","H04N","H04N19/00","H04N19/70","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H","H04","H04N","H04N19/00","H04N19/10","H04N19/169","H04N19/17","H04N19/172","H","H04","H04N","H04N19/00","H04N19/10","H04N19/169","H04N19/17","H04N19/174","H","H04","H04N","H04N19/00","H04N19/30","H","H04","H04N","H04N19/00","H04N19/44","H","H04","H04N","H04N19/00","H04N19/46","H","H04","H04N","H04N19/00","H04N19/46","H04N19/463","H","H04","H04N","H04N19/00","H04N19/60","H04N19/61","H","H04","H04N","H04N19/00","H04N19/90"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025287336A1/en"},{"publication_number":"AU2025287342A1","title":"FCVCM complexity limits","abstract":"FCVCM COMPLEXITY LIMITS A method of decoding tensors for content from a bitstream. The method determines a network topology. A required network complexity is determined from the network topology. A network complexity indication is determined for the bitstream. The tensors for the content are decoded from the bitstream using the determined network topology, depending on the required network complexity and the network complexity indication.","assignee":"Canon Inc","inventors":["Christopher James ROSEWARNE"],"publication_date":"2026-01-22","filing_date":"2025-12-24","priority_date":"2023-05-19","cpc_codes":["H","H04","H04N","H04N19/00","H04N19/70","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06T","G06T9/00","G06T9/002","H","H04","H04N","H04N19/00","H04N19/10","H04N19/134","H04N19/156","H","H04","H04N","H04N19/00","H04N19/44","H","H04","H04N","H04N19/00","H04N19/60","H04N19/61","H","H04","H04N","H04N19/00","H04N19/90"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025287342A1/en"},{"publication_number":"KR20260009951A","title":"Device and method of generating and predicting learning models for cmf and object prediction","abstract":"본 발명은 CMF 및 객체 추정용 학습모델 생성과 추정 장치 및 그 방법에 관한 것으로, 객체 이미지를 기반으로 객체의 색깔(color), 소재(material) 및 마감(finishing)과 관련한 CMF 정보를 추정하고, 상기 객체 이미지와 CMF 정보를 기반으로 상기 객체의 레이블을 추정하는 학습모델 생성 장치 및 그 방법과 이를 이용한 CMF 및 객체 추정 장치 및 그 방법에 관한 것이다. The present invention relates to a device and method for generating and estimating a learning model for CMF and object estimation, and more particularly, to a device and method for generating a learning model for estimating CMF information related to the color, material, and finishing of an object based on an object image, and estimating a label of the object based on the object image and CMF information, and to a device and method for estimating a CMF and object using the same.","assignee":"남서울대학교 산학협력단","inventors":["이수진"],"publication_date":"2026-01-20","filing_date":"2026-01-09","priority_date":"2023-04-05","cpc_codes":["A","A61","A61F","A61F9/00","A61F9/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G09","G09B","G09B21/00","G09B21/001","G09B21/006","G","G10","G10L","G10L13/00","G10L13/02"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260009951A/en"},{"publication_number":"KR20260009791A","title":"Apparatus and method for analyzing pattern of herd behavior of animal based on image","abstract":"영상 기반 군집 개체들의 군집 행동 패턴 분석 장치는 영상 기반 군집 개체의 군집 패턴 분석 프로그램이 저장된 메모리; 및 메모리에 저장된 프로그램을 실행하는 프로세서를 포함하며, 프로그램은, 적어도 하나 이상의 군집 개체가 촬영된 입력 영상을 기반으로 군집 개체의 에지 이미지를 검출하는 영상 전처리를 수행하고, 에지 이미지를 군집 패턴 분석 모델에 입력하여 군집 개체의 패턴 정보를 검출하고, 군집 패턴 분석 모델은 각 군집 개체의 에지 이미지를 포함하는 학습 데이터를 이용하여 학습된 모델로서, 입력 영상을 기준으로 군집 개체의 윤곽 또는 윤곽의 내부 패턴에 의해 결정되는 시각적 형태를 나타내는 군집 개체의 패턴 정보를 출력한다. An image-based cluster behavior pattern analysis device includes a memory in which an image-based cluster pattern analysis program for cluster objects is stored; and a processor executing the program stored in the memory, wherein the program performs image preprocessing to detect an edge image of a cluster object based on an input image in which at least one cluster object is photographed, and inputs the edge image to a cluster pattern analysis model to detect pattern information of the cluster object, and the cluster pattern analysis model is a model learned using learning data including the edge image of each cluster object, and outputs pattern information of the cluster object representing a visual form determined by an outline of the cluster object or an internal pattern of the outline based on the input image.","assignee":"서울대학교산학협력단","inventors":["태주호","정규진","이유빈","김예완","임재욱","배동휘","오은서","김우택"],"publication_date":"2026-01-20","filing_date":"2025-12-30","priority_date":"2022-05-23","cpc_codes":["A","A01","A01K","A01K29/00","A01K29/005","G","G06","G06V","G06V20/00","G06V20/50","G06V20/52","G06V20/53","G","G01","G01J","G01J5/00","G01J5/48","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G06","G06T","G06T7/00","G06T7/90","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06V","G06V40/00","G06V40/20"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260009791A/en"},{"publication_number":"CN121364968A","title":"Computer system fault diagnosis method and system for multi-mode sensing data","abstract":"The invention discloses a computer system fault diagnosis method and system of multi-mode perception data, the method comprises the steps of carrying out unified vectorization representation on the multi-mode perception data of a computer system, carrying out vectorization feature generation by utilizing expert networks of a plurality of channels respectively aiming at the data of each mode in the multi-mode perception data, cascading a shared routing network and a mode routing network of a corresponding mode according to the input data mode to calculate the trust degree of the data of one mode to different expert networks, and carrying out weighted summation on the feature output of each expert network based on the trust degree to obtain the feature representation of the data mode; and sending the characteristics obtained after the characteristic confusion into a classification network to obtain a fault diagnosis result of the computer system. The invention aims to comprehensively utilize multi-mode sensing data to realize unified fault diagnosis of a computer system so as to solve the inconsistent problem when independent fault diagnosis of different modes is carried out.","assignee":"National University of Defense Technology","inventors":["周桐庆","袁远","李志星","邢建英","谢徐超","张根","宋振龙","黎铁军","魏登萍","蔡志平","王承禹","唐滔","吕方旭","邓增","吴振伟"],"publication_date":"2026-01-20","filing_date":"2025-12-23","priority_date":"2025-12-23","cpc_codes":["G","G06","G06F","G06F11/00","G06F11/07","G06F11/0703","G06F11/079","G","G06","G06F","G06F11/00","G06F11/07","G06F11/0703","G06F11/0766","G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2431","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121364968A/en"},{"publication_number":"JP2026009317A","title":"Information processing device, information processing method, and program","abstract":"【課題】認識タスクを行う機械学習モデルについて、特定の事例についての認識精度の改善を効率的に行う。 【解決手段】複数の階層からなる階層的構造を有し、入力されたデータ中の認識対象の認識に用いる機械学習モデルの学習を行う。入力データと、入力データについての機械学習モデルからの出力の正解を示すデータと、を取得する。入力データの特定のドメインについての機械学習モデルからの出力の正解を示すデータと、入力データに対する機械学習モデルの中間層のうち少なくとも１つの出力との誤差に基づいて、機械学習モデルの学習を行う。 【選択図】図２ The present invention provides a method for efficiently improving the recognition accuracy of a machine learning model that performs a recognition task for a specific case. [Solution] A machine learning model having a hierarchical structure consisting of multiple layers is trained and used to recognize a recognition target in input data. The input data and data indicating the correct answer to the output from the machine learning model for the input data are acquired. The machine learning model is trained based on the error between the data indicating the correct answer to the output from the machine learning model for a specific domain of the input data and at least one output of the intermediate layer of the machine learning model for the input data. [Selected Figure] Figure 2","assignee":"Canon Inc","inventors":["敬正 角田"],"publication_date":"2026-01-19","filing_date":"2025-11-04","priority_date":"2021-05-14","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/7715","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/776","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442"],"country":"JP","kind":"application","source_url":"https://patents.google.com/patent/JP2026009317A/en"},{"publication_number":"JP2026009150A","title":"Information processing system, information processing method, and program","abstract":"【課題】コンテンツにより生じる値の目標に応じて、コンテンツ生成のための設定要素の補正に関する情報を出力することが可能な情報処理装置、情報処理方法、およびプログラムを提供する。 【解決手段】コンテンツを生成するために設定される１以上の設定要素の情報に基づいて、前記コンテンツにより生じる値を推定する処理と、前記推定した推定値を目標値と比較する処理と、前記比較の結果に基づいて、前記設定要素の補正に関する補正情報を出力する処理と、を行う制御部を備える、情報処理装置。 【選択図】図１ An information processing device, an information processing method, and a program are provided that are capable of outputting information relating to the correction of setting elements for content generation in accordance with a target value generated by the content. [Solution] An information processing device comprising a control unit that performs the following processes: estimating a value generated by content based on information on one or more setting elements that are set to generate the content; comparing the estimated value with a target value; and outputting correction information regarding correction of the setting elements based on the results of the comparison. [Selected Figure] Figure 1","assignee":"Sony Corp; Sony Group Corp","inventors":["彰吾 木村","奈央 大和","麗子 桐原"],"publication_date":"2026-01-19","filing_date":"2025-10-17","priority_date":"2021-07-21","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0283","G","G06","G06F","G06F16/00","G06F16/90","G06F16/907","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/103","G","G06","G06Q","G06Q50/00","G06Q50/10"],"country":"JP","kind":"application","source_url":"https://patents.google.com/patent/JP2026009150A/en"},{"publication_number":"JP2026009131A","title":"Image processing method, image processing device, and program","abstract":"【課題】 様々なサイズの画像に対して、機械学習モデルを用いた画像処理を高い精度で行うことが可能な画像処理方法を提供する。 【解決手段】 コンピュータを用いて実行される画像処理方法であって、第１の画像を分割し、該第１の画像よりも小さい複数の第２の画像を生成する第１の工程Ｓ２０４と、複数の第２の画像を機械学習モデルに入力することで、複数の第３の画像を生成する第２の工程Ｓ２０５とを有し、複数の第２の画像はそれぞれ、隣接する第２の画像と重複している重複領域を含み、複数の第２の画像は、第１の画素数の重複領域を含む画像と、該第１の画素数と異なる第２の画素数の重複領域を含む画像とを含み生成されること。 【選択図】 図６ An image processing method is provided that is capable of performing image processing using a machine learning model with high accuracy on images of various sizes. [Solution] An image processing method executed using a computer, comprising a first step S204 of dividing a first image and generating a plurality of second images smaller than the first image, and a second step S205 of generating a plurality of third images by inputting the plurality of second images into a machine learning model, wherein each of the plurality of second images includes an overlapping region that overlaps with an adjacent second image, and the plurality of second images generated include an image including an overlapping region with a first number of pixels and an image including an overlapping region with a second number of pixels different from the first number of pixels. [Selected figure] Figure 6","assignee":"Canon Inc","inventors":["良範 木村"],"publication_date":"2026-01-19","filing_date":"2025-10-06","priority_date":"2024-06-27","cpc_codes":["G","G06","G06T","G06T5/00","G06T5/73","G","G06","G06T","G06T11/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06T","G06T11/00","G06T11/40","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G06","G06T","G06T7/00","G06T7/10","G06T7/187","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10004","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20021","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20112"],"country":"JP","kind":"application","source_url":"https://patents.google.com/patent/JP2026009131A/en"},{"publication_number":"KR102914851B1","title":"customized phototherapy-linked skin care system and method based on multi-light source photography and AI analysis","abstract":"본 발명은 디지털 헬스케어 기술과 인공지능(AI) 기반 의료 영상 분석 기술이 융합된 개인 맞춤형 피부 관리 솔루션에 관한 것이다. 보다 상세하게는, 일반광, 편광, 자외선(UV)광을 포함하는 다중 광원을 이용하여 사용자의 피부를 촬영하고, 획득된 다차원 영상 정보를 인공지능 모델로 정밀하게 분석하여 모공, 색소, 주름, 유분 등 핵심 피부 지표를 정량화한다. 또한, 상기 분석 결과를 바탕으로 개인의 피부 상태에 최적화된 LED(발광 다이오드) 광 치료(Photonic Therapy) 프로토콜을 자동으로 생성하고, 사용자의 위치 정보와 연계하여 인근 전문 피부과 또는 관리샵에 분석 데이터를 전송하는 통합적인 온-오프라인 연계 피부 관리 시스템 및 그 운영 방법에 관한 것이다. The present invention relates to a personalized skin care solution that combines digital healthcare technology with artificial intelligence (AI)-based medical image analysis technology. More specifically, the solution captures a user's skin using multiple light sources, including normal light, polarized light, and ultraviolet (UV) light. The acquired multidimensional image data is then precisely analyzed using an AI model to quantify key skin indicators such as pores, pigmentation, wrinkles, and sebum. In addition, the present invention relates to an integrated on-offline linked skin care system and its operation method, which automatically generates an LED (light-emitting diode) photonic therapy protocol optimized for an individual's skin condition based on the above analysis results and transmits the analysis data to a nearby specialized dermatology clinic or management shop in conjunction with the user's location information.","assignee":"정승훈","inventors":["정승훈"],"publication_date":"2026-01-19","filing_date":"2025-09-18","priority_date":"2025-09-18","cpc_codes":["G","G16","G16H","G16H20/00","G16H20/40","A","A61","A61B","A61B5/00","A61B5/44","A61B5/441","A","A61","A61N","A61N5/00","A61N5/06","A61N5/0613","A61N5/0616","F","F21","F21V","F21V33/00","F21V33/0064","F21V33/0068","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G16","G16H","G16H30/00","G16H30/20","G","G16","G16H","G16H30/00","G16H30/40","G","G16","G16H","G16H40/00","G16H40/20","G","G16","G16H","G16H40/00","G16H40/60","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/30"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102914851B1/en"},{"publication_number":"KR102914722B1","title":"Method for depth map enhancement and system therefor","abstract":"깊이맵 보강 방법 및 그 시스템이 제공된다. 몇몇 실시예들에 따른 깊이맵 보강 방법은, 3D 모델링 툴이 제공하는 가상의 공간에 하나 이상의 전경 객체와 하나 이상의 배경 요소를 배치하여 3D 장면을 구성하는 단계, 3D 모델링 툴의 렌더러를 통해 하나 이상의 전경 객체에 관한 제1 깊이맵 샘플을 생성하는 단계, 3D 장면에 관한 제2 깊이맵 샘플을 생성하는 단계 - 제2 깊이맵 샘플은 정답(ground truth) 깊이맵으로 지정되고, 제1 깊이맵 샘플보다 하나 이상의 배경 요소에 대한 깊이 정보를 더 포함함 - 및 제1 깊이맵 샘플과 제2 깊이맵 샘플을 기초로 딥러닝 모델의 트레이닝셋을 준비하는 단계를 포함할 수 있다. 이러한 방법에 다르면, 깊이 정보를 보강하는 딥러닝 모델을 위한 고품질의 트레이닝셋이 용이하게 준비될 수 있다. A depth map augmentation method and system thereof are provided. The depth map augmentation method according to some embodiments may include the steps of composing a 3D scene by arranging one or more foreground objects and one or more background elements in a virtual space provided by a 3D modeling tool, generating a first depth map sample for the one or more foreground objects through a renderer of the 3D modeling tool, generating a second depth map sample for the 3D scene, wherein the second depth map sample is designated as a ground truth depth map and includes more depth information for one or more background elements than the first depth map sample, and preparing a training set for a deep learning model based on the first depth map sample and the second depth map sample. According to this method, a high-quality training set for a deep learning model augmenting depth information can be easily prepared.","assignee":"주식회사 엔닷라이트","inventors":["문종보","김선태"],"publication_date":"2026-01-19","filing_date":"2025-08-25","priority_date":"2025-04-03","cpc_codes":["G","G06","G06T","G06T5/00","G06T5/50","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06T","G06T15/00","G06T15/10","G06T15/20","G06T15/205","G","G06","G06T","G06T19/00","G06T19/20","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G06","G06T","G06T7/00","G06T7/10","G06T7/194","G","G06","G06T","G06T7/00","G06T7/50","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102914722B1/en"},{"publication_number":"CA3281593A1","title":"Systems and methods for automated prediction of insights for vendor product roadmaps","abstract":"Computerized systems and methods are described for generating and optimizing vendor product roadmaps using predictive insights. Leveraging a Real-Time Data Mesh (RTDM) module, data from diverse sources including market trends, customer feedback, and technological advancements is aggregated and standardized. An Analytics and Machine- Learning (AAML) module analyzes this data to generate predictive insights, facilitating adjustments to existing product roadmaps. Dynamic adjustments are made using a roadmap optimization module, with communication facilitated through a Single Pane of Glass (SPoG) user interface (UI). Scenario analysis, decision-support systems, and continuous monitoring improve strategic decision-making.","assignee":"Ingram Micro Inc","inventors":["Sanjib Sahoo"],"publication_date":"2026-01-19","filing_date":"2025-07-30","priority_date":"2024-08-02","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0202","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06315","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087"],"country":"CA","kind":"application","source_url":"https://patents.google.com/patent/CA3281593A1/en"},{"publication_number":"KR20260009312A","title":"Image encoding method, image encoding device, image decoding method, image decoding device, method for transmitting bitstream, and recording medium storing bitstream","abstract":"실시예들에 따른 방법은 픽처들을 위한 SEI(Supplemental Enhancement Information) 메시지를 획득하는 단계; 및 픽처들을 디코딩하는 단계; 를 포함하고, SEI 메시지는 SEI 메시지들의 타입들의 그룹에 대한 프로세싱 오더 정보를 포함할 수 있다. 실시예들에 따른 방법은 픽처들을 위한 SEI(Supplemental Enhancement Information) 메시지를 유도하는 단계; 및 픽처들을 인코딩하는 단계; 를 포함하고, SEI 메시지는 SEI 메시지들의 타입들의 그룹에 대한 프로세싱 오더 정보를 포함할 수 있다. A method according to embodiments includes the steps of obtaining a Supplemental Enhancement Information (SEI) message for pictures; and decoding the pictures, wherein the SEI message may include processing order information for a group of types of SEI messages. A method according to embodiments includes the steps of deriving a Supplemental Enhancement Information (SEI) message for pictures; and encoding the pictures, wherein the SEI message may include processing order information for a group of types of SEI messages.","assignee":"엘지전자 주식회사","inventors":["탄헨드리","임재현","김철근","남정학","이장원","김승환"],"publication_date":"2026-01-19","filing_date":"2025-07-09","priority_date":"2024-07-09","cpc_codes":["H","H04","H04N","H04N19/00","H04N19/70","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/119","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/129","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/13","H","H04","H04N","H04N19/00","H04N19/10","H04N19/169","H04N19/17","H04N19/172","H","H04","H04N","H04N19/00","H04N19/10","H04N19/169","H04N19/184","H","H04","H04N","H04N19/00","H04N19/46","H","H04","H04N","H04N19/00","H04N19/60","H04N19/62"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260009312A/en"},{"publication_number":"KR20260009233A","title":"Method for encoding/decoding multi-scale feature","abstract":"본 개시에 따른 멀티 스케일 특징 부호화 방법은, 멀티 스케일 특징을 단일 스케일 특징으로 변환하는 단계; 상기 단일 스케일 특징에 대해 채널 절삭을 수행하는 단계; 채널 절삭된 단일 스케일 특징에 대해 채널 패킹을 수행하는 단계; 및 상기 채널 패킹에 의해 획득된 2D 프레임을 부호화하는 단계를 포함할 수 있다. 이때, 상기 채널 절삭을 통해, 상기 단일 스케일 특징의 특징 채널들 중 적어도 하나가 제거될 수 있다. A multi-scale feature encoding method according to the present disclosure may include the steps of: converting a multi-scale feature into a single-scale feature; performing channel pruning on the single-scale feature; performing channel packing on the channel-pruned single-scale feature; and encoding a 2D frame obtained by the channel packing. At this time, at least one of the feature channels of the single-scale feature may be removed through the channel pruning.","assignee":"한국전자통신연구원; 한국항공대학교산학협력단","inventors":["정세윤","김연희","이주영","강정원","김재곤","한규웅","김동하"],"publication_date":"2026-01-19","filing_date":"2025-07-08","priority_date":"2024-07-10","cpc_codes":["H","H04","H04N","H04N19/00","H04N19/10","H04N19/134","H04N19/136","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","H","H04","H04N","H04N19/00","H04N19/20"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260009233A/en"},{"publication_number":"KR102912751B1","title":"Solar power generation prediction system using artificial intelligence","abstract":"본 발명은 딥러닝 모델 기반 태양광 발전율 예측장치에 있어서, 특히 발전율 예측 모델 성능의 저하를 해결하기 위해 기상 유사도를 통해 유사한 지역의 과거 데이터를 활용하는 딥러닝 모델 기반 태양광 발전율 예측장치에 관한 것으로, 과거에 측정된 발전량 정보가 풍부한 여러 지역으로부터 기상정보와 태양광 발전량 정보 등을 수집하는 데이터 수집부(100)와; 발전량을 설비용량으로 나눈 발전율 값을 활용하여 표준 정규화를 통해 데이터 수집부(100)가 수집한 데이터를 전처리 하는 데이터 전처리부(200)와; 상기 데이터 전처리부(200)의 결과물로 출력된 발전율 데이터, 기상관측 데이터, 대기질 측정 데이터를 근거로 태양광 발전율 딥러닝 모델을 각각 학습시키는 딥러닝 모델 학습부(300)와; 상기 딥러닝 모델 학습부를 통해 학습된 데이터를 근거로 현재 태양광 패널에 대한 고장 예측 판단을 하는 태양광 패널 에러상태 예측 및 판단부(1000)를 포함하여 이루어지는 것이 특징이다. The present invention relates to a solar power generation rate prediction device based on a deep learning model, and more particularly, to a solar power generation rate prediction device based on a deep learning model that utilizes past data of similar regions through weather similarity to solve the degradation of the performance of the generation rate prediction model. It is characterized by including a data collection unit (100) that collects weather information and solar power generation information from various regions with abundant power generation information measured in the past; a data preprocessing unit (200) that preprocesses the data collected by the data collection unit (100) through standard normalization using a power generation rate value obtained by dividing the power generation by the facility capacity; a deep learning model learning unit (300) that trains a solar power generation rate deep learning model based on the power generation rate data, weather observation data, and air quality measurement data output as results of the data preprocessing unit (200); and a solar panel error status prediction and judgment unit (1000) that predicts and determines a failure of the current solar panel based on the data learned through the deep learning model learning unit.","assignee":"주식회사 효성에너지팜","inventors":["이도훈"],"publication_date":"2026-01-19","filing_date":"2025-07-04","priority_date":"2025-07-04","cpc_codes":["H","H02","H02S","H02S50/00","G","G01","G01R","G01R19/00","G01R19/0046","G01R19/0053","G","G01","G01R","G01R19/00","G01R19/10","G","G01","G01R","G01R19/00","G01R19/12","G","G01","G01R","G01R19/00","G01R19/165","G","G01","G01W","G01W1/00","G01W1/02","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G08","G08B","G08B21/00","G08B21/18","H","H02","H02S","H02S30/00","H","H02","H02S","H02S40/00","H02S40/40","H02S40/42","H02S40/425"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102912751B1/en"},{"publication_number":"JP2026009077A","title":"Lane boundary intersection detection method, device, equipment, and storage medium","abstract":"【課題】インテリジェント運転技術の分野に関する車線境界線交差点の検出方法、装置、機器及び記憶媒体を開示する。 【解決手段】当該方法は、自車が走行中に収集した検出用画像を決定するステップと、車線境界線交差点予測モデルに基づいて検出用画像を処理して、車線境界線交差点種類特徴マップを得るステップと、車線境界線交差点種類特徴マップに基づいて、検出用画像中の車線境界線交差点の座標及び種類を決定するステップとを含む。本開示の技術案にて提供される車線境界線交差点を検出する検出精度は、車線境界線交差点予測モデルの精度に依存し、車線境界線検出モデルの精度に制限されない。かつ、車線境界線交差点予測モデルを採用して車線境界線交差点を検出する方式は、遠すぎる位置及び近すぎる位置の車線境界線交差点の画素点数の影響を受けないため、車線境界線交差点検出の有効性を向上させることができる。 【選択図】図１ A method, apparatus, device and storage medium for detecting lane boundary intersections in the field of intelligent driving technology are disclosed. [Solution] The method includes the steps of determining detection images collected by the vehicle while it is traveling, processing the detection images based on a lane boundary intersection prediction model to obtain a lane boundary intersection type feature map, and determining the coordinates and types of lane boundary intersections in the detection images based on the lane boundary intersection type feature map. The detection accuracy of the lane boundary intersection detection provided by the technical solution disclosed herein depends on the accuracy of the lane boundary intersection prediction model, but is not limited by the accuracy of the lane boundary detection model. Furthermore, the method of detecting lane boundary intersections using a lane boundary intersection prediction model is not affected by the number of pixels at lane boundary intersections that are too far or too close, thereby improving the effectiveness of lane boundary intersection detection. [Selected Figure] Figure 1","assignee":"南京地平▲線▼信息技▲術▼有限公司","inventors":["ボルイ ツァオ","インキアン ツァオ"],"publication_date":"2026-01-19","filing_date":"2025-07-03","priority_date":"2024-07-03","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/50","G06V20/56","G06V20/58","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06V","G06V10/00","G06V10/20","G06V10/22","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/50","G06V20/56","G06V20/588"],"country":"JP","kind":"application","source_url":"https://patents.google.com/patent/JP2026009077A/en"},{"publication_number":"KR20260008711A","title":"Method and device to prevent invasion of privacy and detect defect on building walls","abstract":"개시된 기술은 사생활 보호를 위한 건물 외벽 하자 검출 방법 및 장치에 관한 것으로, 검출기가 건물 외벽에 대한 영상을 수신하는 단계; 상기 검출기가 상기 영상에서 비식별 대상을 검출하는 단계; 상기 검출기가 상기 영상에서 상기 비식별 대상이 포함된 영역을 제거하여 이미지 패치(Patch)를 생성하는 단계; 및 상기 검출기가 상기 이미지 패치를 검출 모델에 입력하여 상기 건물 외벽에 발생한 하자를 검출하는 단계;를 포함한다. The disclosed technology relates to a method and device for detecting a defect in a building exterior wall for privacy protection, comprising: a step in which a detector receives an image of a building exterior wall; a step in which the detector detects an unidentifiable object in the image; a step in which the detector removes an area including the unidentifiable object from the image to generate an image patch; and a step in which the detector inputs the image patch into a detection model to detect a defect occurring in the building exterior wall.","assignee":"연세대학교 산학협력단; 한양대학교 에리카산학협력단","inventors":["김하영","이기수","이상효","신현규","윤종현","김종훈"],"publication_date":"2026-01-16","filing_date":"2025-12-26","priority_date":"2021-10-01","cpc_codes":["G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G","G01","G01B","G01B11/00","G01B11/16","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T3/00","G06T3/40","G06T3/4038","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G01N2021/8887"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260008711A/en"},{"publication_number":"KR20260008705A","title":"A System for Supporting a Personal Safety and a Sailing Safety Based on an Artificial Intelligence and a Supporting Method with the Same","abstract":"본 발명은 인공지능 기반 해상 인명 안전 및 운항 안전 지원 시스템 및 그에 의한 지원 방법에 관한 것이다. 인공지능 기반 해상 인명 안전 및 운항 안전 지원 시스템은 선박 내부 또는 선박 외부의 미리 정해진 영역의 영상을 연속적으로 획득하는 영상 획득 모듈(11); 영상 획득 모듈(11)에 의하여 획득된 영상 데이터로부터 감시 객체를 추출하고, 추출된 감시 객체를 분석하여 위험 상황을 판단하는 신호 처리 모듈(12); 신호 처리 모듈(12)에서 판단된 위험 상황을 외부에 표시하는 영상 전시 모듈(13); 및 위험 상황을 선박 안전 관련 기관에 무선 통신을 통하여 전송하는 통신 모듈(14)을 포함한다. The present invention relates to an artificial intelligence-based maritime human life safety and navigation safety support system and a support method thereof. The artificial intelligence-based maritime human life safety and navigation safety support system comprises an image acquisition module (11) that continuously acquires images of a predetermined area inside or outside a ship; a signal processing module (12) that extracts surveillance objects from image data acquired by the image acquisition module (11) and analyzes the extracted surveillance objects to determine a dangerous situation; an image display module (13) that externally displays the dangerous situation determined by the signal processing module (12); and a communication module (14) that transmits the dangerous situation to a ship safety-related organization via wireless communication.","assignee":"주식회사 산엔지니어링","inventors":["김효성","정다은","이태석","조대영","정경진","이주형"],"publication_date":"2026-01-16","filing_date":"2025-12-24","priority_date":"2020-11-06","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26","G06Q50/265","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06Q","G06Q50/00","G06Q50/40","G","G06","G06V","G06V40/00","G06V40/20","G","G08","G08B","G08B25/00","G08B25/01","G08B25/10","G","G08","G08B","G08B27/00","G08B27/001","H","H04","H04N","H04N7/00","H04N7/18","H04N7/181"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260008705A/en"},{"publication_number":"CN121351654A","title":"An optimized method and apparatus for cold plates in containerized energy storage power stations","abstract":"本发明涉及电化学储能系统热管理技术领域，公开了一种集装箱式储能电站冷板的优化方法及装置，该方法包括：建立多物理场耦合仿真模型；获取目标集装箱式储能电站冷板中分形流道的几何特征，将分形流道的几何特征输入多物理场耦合仿真模型中，得到性能指标；构建物理引导的机器学习代理模型；将分形流道的几何特征和性能指标作为样本数据，利用样本数据对物理引导的机器学习代理模型进行训练，得到训练后的机器学习代理模型；对训练后的机器学习代理模型进行多目标优化和拓扑调整，得到最优集装箱式储能电站冷板结构。本发明解决了因电池模组在集装箱内空间分布导致的热环境差异，显著提升了电化学储能系统的热管理效率与安全性。 This invention relates to the field of thermal management technology for electrochemical energy storage systems, and discloses an optimization method and apparatus for cold plates in containerized energy storage power stations. The method includes: establishing a multiphysics coupled simulation model; obtaining the geometric features of fractal flow channels in the target containerized energy storage power station's cold plate, inputting the geometric features of the fractal flow channels into the multiphysics coupled simulation model to obtain performance indicators; constructing a physics-guided machine learning proxy model; using the geometric features and performance indicators of the fractal flow channels as sample data, training the physics-guided machine learning proxy model using the sample data to obtain a trained machine learning proxy model; and performing multi-objective optimization and topology adjustment on the trained machine learning proxy model to obtain the optimal cold plate structure for the containerized energy storage power station. This invention solves the problem of thermal environment differences caused by the spatial distribution of battery modules within the container, significantly improving the thermal management efficiency and safety of electrochemical energy storage systems.","assignee":"Xian Jiaotong University; Sungrow Power Supply Co Ltd; Huadian Electric Power Research Institute Co Ltd","inventors":["张海珍","严新荣","李印实","周俭杰","谢玉荣","邓睿锋","陈桥","刘丽丽","袁江伟","牟敏","杨皓杰","焦君昊","丁可"],"publication_date":"2026-01-16","filing_date":"2025-12-22","priority_date":"2025-12-22","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N20/00","G","G06","G06F","G06F2111/00","G06F2111/04","G","G06","G06F","G06F2111/00","G06F2111/06","G","G06","G06F","G06F2119/00","G06F2119/08","Y","Y02","Y02E","Y02E60/00","Y02E60/10"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121351654A/en"},{"publication_number":"CN121354940A","title":"A spinal cord injury status assessment system integrating pathological data and clinical characteristics","abstract":"The invention discloses a spinal cord injury state assessment system integrating pathological data and clinical features, which comprises a multi-mode data synchronous acquisition module, an injury feature collaborative extraction module, an individuation state quantitative assessment module, an adaptive rehabilitation strategy generation module and an adaptive rehabilitation strategy generation module, wherein the multi-mode data synchronous acquisition module is used for multi-mode original data flow, the injury feature collaborative extraction module is used for extracting structural injury features and functional injury features through a double-channel deep learning network based on an attention mechanism and calculating an association weight matrix, the individuation state quantitative assessment module is used for generating injury state vectors based on the weight matrix fusion features and outputting a comprehensive assessment report through a pre-training model, and the adaptive rehabilitation strategy generation module is used for generating an individuation stage type rehabilitation treatment scheme according to the comprehensive assessment report and decomposing the individuation stage type rehabilitation treatment scheme into an executable instruction sequence. The invention has the following advantages and effects that the accurate assessment of the spinal cord injury state and the accurate formulation of the rehabilitation strategy are realized by quantifying the personalized mapping relation between the spinal cord structural injury and the physical dysfunction.","assignee":"First Affiliated Hospital of Nanchang University","inventors":["王立超","蓝淳愉","张科","付方智"],"publication_date":"2026-01-16","filing_date":"2025-12-22","priority_date":"2025-12-22","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/30","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H50/00","G16H50/70"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121354940A/en"},{"publication_number":"CN121351977A","title":"A data analysis method, apparatus and equipment","abstract":"本申请实施例提供一种数据分析方法、装置及设备，涉及信息安全技术领域，该方法包括：获取不同来源的原始安全数据，并生成各原始安全数据分别对应的描述文本；不同来源的原始安全数据包括多模态的安全数据；针对每个描述文本，基于第一大语言模型识别描述文本中包括的多个实体以及各实体之间的语义连接关系，并基于多个实体以及实体之间的语义连接关系生成语义图谱子图；语义图谱子图中的节点用于表征实体，节点之间的连线用于表征实体之间的语义连接关系；对各描述文本分别对应的语义图谱子图进行融合处理，得到目标关系图谱，有效增强了数据分析的泛化能力、语义理解能力和跨模态集成能力。 This application provides a data analysis method, apparatus, and device, relating to the field of information security technology. The method includes: acquiring raw security data from different sources and generating descriptive text corresponding to each raw security data; the raw security data from different sources includes multimodal security data; for each descriptive text, identifying multiple entities included in the descriptive text and the semantic connection relationships between the entities based on a first major language model, and generating a semantic graph subgraph based on the multiple entities and the semantic connection relationships between the entities; nodes in the semantic graph subgraph are used to represent entities, and the connections between nodes are used to represent the semantic connection relationships between entities; and the semantic graph subgraphs corresponding to each descriptive text are fused to obtain a target relationship graph, effectively enhancing the generalization ability, semantic understanding ability, and cross-modal integration ability of data analysis.","assignee":"Nsfocus Technologies Group Co Ltd","inventors":["顾杜娟"],"publication_date":"2026-01-16","filing_date":"2025-12-22","priority_date":"2025-12-22","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/289","G06F40/295","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121351977A/en"},{"publication_number":"CN121353563A","title":"Data expansion methods, apparatus, computer equipment and storage media","abstract":"本发明涉及自动驾驶技术领域，公开了数据扩充方法、装置、计算机设备及存储介质，包括获取第一场景描述；基于所述第一场景描述从预设地图数据集中确定匹配的目标地图区域，并基于第一场景描述、目标地图区域以及预设交通流生成模型，生成动态目标的三维轨迹信息；基于三维轨迹信息以及目标车辆进行坐标转换，生成与目标车辆对应的三维边界框结构化信息和车道线结构化信息；对三维边界框结构化信息和车道线结构化信息进行处理，生成扩充数据。本发明通过自然语言形式的第一场景描述匹配目标地图区域并生成动态目标三维轨迹，经坐标转换得到结构化信息后生成多视角且连续帧的场景数据，降低自动驾驶数据的获取成本，进而提升模型的泛化能力。 This invention relates to the field of autonomous driving technology and discloses a data augmentation method, apparatus, computer equipment, and storage medium. The method includes: acquiring a first scene description; determining a matching target map region from a preset map dataset based on the first scene description; generating three-dimensional trajectory information of a dynamic target based on the first scene description, the target map region, and a preset traffic flow generation model; performing coordinate transformation based on the three-dimensional trajectory information and the target vehicle to generate three-dimensional bounding box structured information and lane line structured information corresponding to the target vehicle; and processing the three-dimensional bounding box structured information and lane line structured information to generate augmented data. This invention matches a target map region with a first scene description in natural language form and generates a dynamic target's three-dimensional trajectory. After coordinate transformation to obtain structured information, it generates multi-view and continuous frame scene data, reducing the cost of acquiring autonomous driving data and thus improving the model's generalization ability.","assignee":"Magic Vision Intelligent Technology Shanghai Co ltd","inventors":["刘国辉","虞正华"],"publication_date":"2026-01-16","filing_date":"2025-12-22","priority_date":"2025-12-22","cpc_codes":["G","G06","G06T","G06T17/00","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2455","G","G06","G06F","G06F16/00","G06F16/20","G06F16/29","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121353563A/en"},{"publication_number":"KR20260008059A","title":"Control system using AI function of automatic water cannon","abstract":"본 발명은 자동방수포의 인공지능 기능을 이용한 제어시스템에 관한 것이다. 이때, 본 발명의 자동방수포는 노즐, 구동 모듈, 카메라, 제어부, 및 원격조종기를 포함한다. 상기 카메라는 화재 현장을 실시간으로 모니터링하고, 제어부는 카메라로부터 받은 영상 데이터를 분석하여 화점을 탐지한다. 화점이 식별되면 제어부는 구동 모듈을 제어하여 노즐을 화점 방향으로 조준하고, 최적의 소화액 분사 패턴을 결정한다. 상기 제어부는 인공지능 알고리즘을 탑재하여 화재 진압 전략을 수립하며, 화재 진압 과정에서 수집한 데이터를 바탕으로 인공지능 학습을 수행하여 성능을 개선한다. 상기 원격조종기를 통해 사용자는 안전한 거리에서 방수포를 제어할 수 있으며, 자동 제어가 불가능한 상황에서도 수동으로 조작할 수 있다. 본 발명에 의한 자동방수포는 인공지능 기능을 활용하여 화재 상황에 능동적으로 대처할 수 있으며, 원격 제어 및 자동/수동 조작 모드를 지원하여 사용자의 안전성과 편의성을 높인다. 또한, 인공지능 학습을 통해 화재 진압 성능을 지속적으로 개선할 수 있다. 이를 통해 본 발명은 화재로 인한 피해를 최소화하고, 화재 진압 작업의 효율성과 안전성을 크게 향상시킬 수 있을 것으로 기대된다. The present invention relates to a control system using the artificial intelligence function of an automatic water gun. At this time, the automatic waterproof gun of the present invention includes a nozzle, a driving module, a camera, a control unit, and a remote control unit. The above camera monitors the fire scene in real time, and the control unit analyzes the image data received from the camera to detect the fire source. Once the fire point is identified, the control unit controls the drive module to aim the nozzle toward the fire point and determine the optimal extinguishing agent spray pattern. The above control unit establishes a fire suppression strategy by installing an artificial intelligence algorithm, and improves performance by performing artificial intelligence learning based on data collected during the fire suppression process. The above remote control allows the user to control the tarpaulin from a safe distance and to operate it manually even in situations where automatic control is not possible. The automatic fire blanket according to the present invention can actively respond to a fire situation by utilizing an artificial intelligence function, and supports remote control and automatic/manual operation modes to enhance user safety and convenience. Additionally, fire suppression performance can be continuously improved through artificial intelligence learning. Through this, it is expected that the present invention will be able to minimize damage caused by fire and greatly improve the efficiency and safety of fire suppression operations.","assignee":"주식회사 케이대응로봇","inventors":["구광민"],"publication_date":"2026-01-15","filing_date":"2025-12-23","priority_date":"2024-04-26","cpc_codes":["A","A62","A62C","A62C37/00","A62C37/36","A62C37/38","A","A62","A62C","A62C27/00","A","A62","A62C","A62C31/00","A62C31/02","A","A62","A62C","A62C99/00","A62C99/009","G","G05","G05G","G05G9/00","G05G9/02","G05G9/04","G05G9/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06V","G06V10/00","G06V10/20","G06V10/255","H","H04","H04N","H04N23/00","H04N23/10","H","H04","H04N","H04N23/00","H04N23/20","H04N23/23"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260008059A/en"},{"publication_number":"KR20260007555A","title":"Device and method for determining anomalies of an object using artificial intelligence model","abstract":"본 발명의 실시예에 따른 컴퓨터가 인공지능 모델을 이용하여 대상물의 이상을 감지하는 방법에 있어서, 상기 컴퓨터의 적어도 하나의 프로세서가, 상기 대상물을 시간 단위에 대응하여 측정한 입력 데이터를 획득하는 단계; 상기 적어도 하나의 프로세서가, 상기 획득된 입력 데이터에 포함된 복수의 동종 시계열 데이터인 제1 데이터를 정규화하는 전처리를 수행하여 제2 데이터를 생성하는 단계; 상기 적어도 하나의 프로세서가, 상기 생성된 제2 데이터를 인공지능 모델에 입력하는 단계; 상기 인공지능 모델에서 출력된 상기 제2 데이터의 복원 데이터를 기초로 이상 스코어가 출력되는 단계; 상기 적어도 하나의 프로세서가, 상기 출력된 이상 스코어와 기 설정된 임계값을 비교하여 상기 대상물의 이상 여부를 결정하는 단계를 포함한다. In accordance with an embodiment of the present invention, a method for detecting an abnormality in an object using an artificial intelligence model by a computer comprises: a step in which at least one processor of the computer acquires input data measuring the object in units of time; a step in which the at least one processor performs preprocessing to normalize first data, which is a plurality of homogeneous time-series data included in the acquired input data, to generate second data; a step in which the at least one processor inputs the generated second data into an artificial intelligence model; a step in which an abnormality score is output based on restored data of the second data output from the artificial intelligence model; and a step in which the at least one processor compares the output abnormality score with a preset threshold value to determine whether the object is abnormal.","assignee":"주식회사 Lg 경영개발원","inventors":["임우형","심예슬","조혜승","윤수희"],"publication_date":"2026-01-14","filing_date":"2025-12-31","priority_date":"2022-10-06","cpc_codes":["G","G01","G01R","G01R31/00","G01R31/36","G01R31/367","G","G01","G01R","G01R31/00","G01R31/36","G01R31/3644","G01R31/3648","G","G01","G01R","G01R31/00","G01R31/36","G01R31/382","G01R31/3842","G","G01","G01R","G01R31/00","G01R31/36","G01R31/392","G","G01","G01R","G01R31/00","G01R31/36","G01R31/396","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260007555A/en"},{"publication_number":"KR20260007548A","title":"Bus Energy System Installed in a Hydrogen-Electric Hybrid Bus that Controls Fuel Cell Operation According to Energy Prediction Results Based on Multiple Prediction Models","abstract":"실시 예들은 수소전기 겸용 버스의 GPS 위치를 결정하도록 구성된 GPS 모듈;서버 및 제3자 네비게이션 시스템과 통신하여 정보를 송수신하는 통신부; 배터리의 속성 정보 및 상기 연료전지의 속성 정보를 측정하는 센서; 및 수소전기 겸용 버스의 전기 에너지를 관리하여 상기 수소전기 겸용 버스가 노선을 따라 주행하도록 제어하는 제어부를 포함하는, 미리 정해진 노선을 주행하는 수소전기 겸용 버스에 설치되는 것으로서, 배터리 및 연료전지를 포함하고, 서버 및 제3자 네비게이션 시스템과 통신하여 배터리, 연료전지 동작을 제어하는 버스 에너지 시스템에 관련된다. Embodiments relate to a bus energy system, which is installed in a hydrogen-electric bus that runs on a predetermined route, including a GPS module configured to determine a GPS location of the hydrogen-electric bus; a communication unit that communicates with a server and a third-party navigation system to transmit and receive information; a sensor that measures property information of a battery and property information of the fuel cell; and a control unit that manages electric energy of the hydrogen-electric bus to control the hydrogen-electric bus to run along the route, the bus energy system including a battery and a fuel cell, and that communicates with a server and a third-party navigation system to control the operation of the battery and the fuel cell.","assignee":"서울버스(주)","inventors":["조준서"],"publication_date":"2026-01-14","filing_date":"2025-12-24","priority_date":"2023-02-20","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/40","B","B60","B60L","B60L53/00","B60L53/60","B","B60","B60L","B60L58/00","B60L58/40","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q50/00","G06Q50/10","B","B60","B60L","B60L2200/00","B60L2200/18","B","B60","B60L","B60L2240/00","B60L2240/60","B","B60","B60L","B60L2260/00","B60L2260/40","B60L2260/50","B60L2260/54","B","B60","B60Y","B60Y2200/00","B60Y2200/90","B60Y2200/91","B","B60","B60Y","B60Y2400/00","B60Y2400/10","B60Y2400/102","B","B60","B60Y","B60Y2400/00","B60Y2400/11","B60Y2400/112","Y","Y02","Y02E","Y02E60/00","Y02E60/30","Y02E60/50","Y","Y02","Y02T","Y02T10/00","Y02T10/60","Y02T10/70","Y","Y02","Y02T","Y02T10/00","Y02T10/60","Y02T10/7072"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260007548A/en"},{"publication_number":"KR20260007540A","title":"Video codec assisted real-time video enhancement using deep learning","abstract":"비디오 코덱 정보에 기반하여 선택적으로 적용되는 딥 러닝을 사용하는 가속화된 비디오 향상에 관련된 기법이 논의된다. 그러한 기법은 낮은 양자화 파라미터 프레임 내에 있는 디코딩된 비 스킵 블록에 선택적으로 딥 러닝 비디오 향상 네트워크를 적용하는 것과, 낮은 양자화 파라미터 프레임 내의 디코딩된 스킵 블록에 대해 딥 러닝 네트워크를 바이패스하는 것과, 높은 양자화 파라미터 프레임에 비 딥 러닝 비디오 향상을 적용하는 것을 포함한다. Techniques for accelerated video enhancement using deep learning that are selectively applied based on video codec information are discussed. Such techniques include selectively applying a deep learning video enhancement network to decoded non-skippable blocks within low quantization parameter frames, bypassing the deep learning network for decoded skipped blocks within low quantization parameter frames, and applying non-deep learning video enhancement to high quantization parameter frames.","assignee":"인텔 코포레이션","inventors":["첸 왕","시민 장","후안 도우","이-젠 치우","상희 이"],"publication_date":"2026-01-14","filing_date":"2025-12-23","priority_date":"2020-06-26","cpc_codes":["G","G06","G06T","G06T9/00","G06T9/002","H","H04","H04N","H04N19/00","H04N19/44","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/251","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06T","G06T3/00","G06T3/40","G06T3/4007","G","G06","G06T","G06T3/00","G06T3/40","G06T3/4046","G","G06","G06T","G06T3/00","G06T3/40","G06T3/4053","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/103","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/124","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/132","H","H04","H04N","H04N19/00","H04N19/10","H04N19/134","H04N19/157","H04N19/159","H","H04","H04N","H04N19/00","H04N19/10","H04N19/169","H04N19/17","H04N19/172","H","H04","H04N","H04N19/00","H04N19/10","H04N19/169","H04N19/17","H04N19/176","H","H04","H04N","H04N19/00","H04N19/10","H04N19/169","H04N19/184","H","H04"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260007540A/en"},{"publication_number":"KR20260007171A","title":"Method for increasing convenience of labeling data by using Oriented Bounding Box","abstract":"본 개시의 일 실시예에 따라 컴퓨팅 장치의 하나 이상의 프로세서에 의해 수행되는, 신경망 모델을 학습시키기 위한 데이터를 획득하는 방법이 개시된다. 상기 방법은, 샘플 이미지를 획득하고, 상기 샘플 이미지에 대한 제 1 사용자 입력을 수신하는 단계; 상기 제 1 사용자 입력이 수신된 샘플 이미지에 대해 제 2 사용자 입력을 수신하는 단계; 상기 제 1 사용자 입력 및 상기 제 2 사용자 입력에 기초하여 상기 샘플 이미지에 대한 제 1 방향을 결정하는 단계; 상기 결정된 제 1 방향에 기초하여 제 3 사용자 입력을 수신하고, 상기 제 1 사용자 입력, 상기 제 2 사용자 입력 및 상기 제 3 사용자 입력에 기초하여 상기 샘플 이미지에 대한 라벨링 영역을 획득하는 단계; 및 상기 라벨링 영역에 기초하여 학습 데이터를 획득하는 단계를 포함할 수 있다. According to one embodiment of the present disclosure, a method for acquiring data for training a neural network model is disclosed, which is performed by one or more processors of a computing device. The method may include the steps of: acquiring a sample image and receiving a first user input for the sample image; receiving a second user input for the sample image for which the first user input has been received; determining a first direction for the sample image based on the first user input and the second user input; receiving a third user input based on the determined first direction and acquiring a labeling region for the sample image based on the first user input, the second user input, and the third user input; and acquiring training data based on the labeling region.","assignee":"주식회사 아이브","inventors":["신혜찬"],"publication_date":"2026-01-13","filing_date":"2025-12-26","priority_date":"2024-06-25","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/70","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06V","G06V10/00","G06V10/20","G06V10/25","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260007171A/en"},{"publication_number":"KR20260007170A","title":"Method and system for providing prescribing informationbased on language model","abstract":"본 발명에 따른 언어 모델 기반의 처방 정보 제공 방법은, 의료진(medical staff) 계정으로 로그인된 의료진 단말기로부터 운동 커리큘럼 추천 요청을 수신하는 단계, 상기 요청에 응답하여, 서버로부터 상기 의료진 계정에 연계된 의료진의 의료진 메타 정보 및 상기 의료진 계정에 연계된 치료 대상 환자(patient)의 환자 메타 정보를 추출하는 단계, 상기 의료진 메타 정보 및 상기 환자 메타 정보를 이용하여, 기 설정된 포맷의 프롬프트 생성하는 단계, 상기 프롬프트를 입력으로 받는 거대 언어 모델(LLM: Large Language Model)을 이용하여, 상기 운동 커리큘럼을 획득하는 단계 및 획득된 상기 운동 커리큘럼을 상기 의료진 단말기에 제공하는 단계를 포함할 수 있다. A method for providing prescription information based on a language model according to the present invention may include the steps of: receiving a request for recommending an exercise curriculum from a medical staff terminal logged in with a medical staff account; extracting, in response to the request, medical staff meta information of a medical staff member linked to the medical staff account and patient meta information of a patient to be treated linked to the medical staff account from a server; generating a prompt in a preset format using the medical staff meta information and the patient meta information; obtaining the exercise curriculum using a Large Language Model (LLM) that receives the prompt as an input; and providing the obtained exercise curriculum to the medical staff terminal.","assignee":"에버엑스 주식회사","inventors":["김병훈","윤찬"],"publication_date":"2026-01-13","filing_date":"2025-12-24","priority_date":"2023-09-27","cpc_codes":["G","G16","G16H","G16H20/00","G16H20/30","A","A63","A63B","A63B24/00","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F16/00","G06F16/30","G06F16/38","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G16","G16H","G16H10/00","G16H10/20","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H40/00","G16H40/20","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/70","G","G16","G16H","G16H80/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260007170A/en"},{"publication_number":"KR20260006506A","title":"A method of predicting the timing of occupancy of rental housing using subscription-related information and a device for this","abstract":"모바일 디바이스에서 실행되는 어플리케이션을 통한 임대주택 청약의 입주 시기 예측 방법이 개시된다. 본 발명의 입주 시기 예측 방법은 특정 임대주택 청약의 예비 순번을 부여받은 사용자로부터 청약 대상이 되는 특정 임대주택 정보 및 예비 순번 정보를 입력받는 단계, 입력된 특정 임대주택 정보에 기초하여, 특정 임대주택의 입주 시기 예측을 위한 청약 관련 정보를 수집하는 단계, 수집된 청약 관련 정보, 사용자의 예비 순번 정보, 기 등록된 과거 임대주택의 청약 관련 정보, 입주 및 퇴거 이력 정보를 비교하여 사용자의 입주 시기를 예측하는 단계 및, 사용자의 모바일 디바이스에 예측된 입주 시기에 대한 예측 정보를 가공하여 제공하는 단계를 포함한다. A method for predicting the move-in time of a rental housing subscription through an application running on a mobile device is disclosed. The method for predicting the move-in time of the present invention includes the steps of receiving, from a user who has been assigned a preliminary turn number for a specific rental housing subscription, information on a specific rental housing subscription subject to the subscription and preliminary turn number information, collecting subscription-related information for predicting the move-in time of the specific rental housing based on the input specific rental housing information, comparing the collected subscription-related information, the user's preliminary turn number information, subscription-related information of previously registered rental housing, and move-in and move-out history information to predict the user's move-in time, and processing and providing the predicted information on the predicted move-in time to the user's mobile device.","assignee":"이상민","inventors":["이상민"],"publication_date":"2026-01-13","filing_date":"2025-12-18","priority_date":"2023-02-06","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/16","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/16","G06Q50/163","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G07","G07C","G07C11/00","G","G07","G07C","G07C11/00","G07C2011/04"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260006506A/en"},{"publication_number":"CN121327100A","title":"A method, equipment, and medium for technical supervision of electrical energy.","abstract":"本发明涉及数据处理技术领域，具体提供了一种电力能源的技术监督问答方法、设备及介质，方法包括：将电力与能源行业的技术监督相关文档构建为结构化的知识库，并对知识库建立混合索引；通过混合索引与重排机制处理用户的查询问题，生成第一候选条款集合和第一答案；将第一候选条款集合和第一答案输入至预先训练好的充分性判定打分器，对第一答案进行回答充分性判定；在第一答案回答不充分时，对用户的查询问题进行补充检索，生成补充答案；将第一答案与补充答案融合，得到第二答案。本发明引入了充分性判定与定向补充，能够自我验证和自我优化，输出具有完整证据链的结构化结果，极大地提升了答案的可信度和直接可用性。 This invention relates to the field of data processing technology, specifically providing a method, device, and medium for technical supervision question answering in the power and energy sector. The method includes: constructing a structured knowledge base from technical supervision-related documents in the power and energy industry, and establishing a hybrid index for the knowledge base; processing user queries through a hybrid index and reordering mechanism to generate a first set of candidate clauses and a first answer; inputting the first set of candidate clauses and the first answer into a pre-trained sufficiency scoring device to determine the sufficiency of the first answer; if the first answer is insufficient, performing supplementary retrieval of the user's query to generate a supplementary answer; and merging the first answer and the supplementary answer to obtain a second answer. This invention introduces sufficiency judgment and targeted supplementation, enabling self-verification and self-optimization, and outputting structured results with a complete chain of evidence, greatly improving the credibility and direct usability of the answers.","assignee":"State Grid Shanxi Electric Power Co ltd Taiyuan Power Supply Branch","inventors":["周焱","赵冉冉","孙宝鑫","赵文敏","侯昱兴","张超林","张宇婧","李伟","谭艳妮"],"publication_date":"2026-01-13","filing_date":"2025-12-18","priority_date":"2025-12-18","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/30","G06F16/31","G06F16/316","G06F16/319","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3344","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F16/00","G06F16/30","G06F16/36","G06F16/367","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121327100A/en"},{"publication_number":"CN121330687A","title":"An image annotation control method","abstract":"The invention discloses an image annotation control method which comprises the steps of generating an intermediate image sequence through an interpolation method according to an initial image and a labeling image of a target task of the initial image, inserting the intermediate image sequence between the corresponding initial image and the labeling image to obtain an image group, splicing the image groups of the same type into the image group sequence according to the type of the target task, forming a training data set by using the training data set and adopting a low-rank adaptation method to train and adjust an autoregressive large-scale visual model, selecting the corresponding image group in the training data set according to the target task to serve as a reference image group, splicing the reference image group with an image to be labeled into an input sequence, inputting the input sequence into the adjusted model to perform visual reasoning, and generating the intermediate image and a labeling result of the corresponding target task of the image to be labeled. The performance of the model in various visual reasoning tasks is improved through interpolation enhanced data construction and progressive reasoning.","assignee":"Zhejiang University ZJU","inventors":["陈为","王鑫阳","郑可成","朱闽峰","吴蔚","翟伟"],"publication_date":"2026-01-13","filing_date":"2025-12-18","priority_date":"2025-12-18","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/70","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06T","G06T17/00","G06T17/20"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121330687A/en"},{"publication_number":"CN121328689A","title":"A method and system for knowledge graph construction and association analysis of fire supervision and communication full-domain data.","abstract":"本发明涉及智慧消防及大数据分析技术领域，公开了一种面向消防监督及通信全域数据的知识图谱构建与关联分析方法及系统。为解决消防数据异构、关联单一的技术难题，本发明提出领域本体驱动的知识图谱自动化构建与深度分析方法。其核心创新在于：采用多源异构数据融合模型统一消防文书、物联感知等多模态数据的语义表示；利用预训练模型进行实体关系联合抽取，高效构建海量三元组的消防知识图谱；并引入图神经网络进行深度关联挖掘与推理，量化评估风险传导路径并预测潜在威胁。本发明将离散的消防数据提升为可计算、可推理的知识体系，实现消防安全态势的深度感知与智能决策。 This invention relates to the fields of smart fire protection and big data analytics, and discloses a method and system for constructing and analyzing knowledge graphs for fire supervision and communication data across the entire domain. To address the technical challenges of heterogeneous fire data and limited correlation, this invention proposes a domain ontology-driven method for automated construction and deep analysis of knowledge graphs. Its core innovations lie in: employing a multi-source heterogeneous data fusion model to unify the semantic representation of multimodal data such as fire documents and IoT sensing data; utilizing a pre-trained model for joint entity relation extraction to efficiently construct a massive triple-based fire knowledge graph; and introducing graph neural networks for deep correlation mining and reasoning to quantitatively assess risk transmission paths and predict potential threats. This invention elevates discrete fire data into a computable and reasonable knowledge system, enabling deep perception and intelligent decision-making regarding fire safety situations.","assignee":"Tezhijia Changsha Iot Technology Co ltd","inventors":["郑妍","王然","赵方强","张婷婷","王斌","张凯明","王承龙","彭统华","周衡","彭拯"],"publication_date":"2026-01-13","filing_date":"2025-12-18","priority_date":"2025-12-18","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/02","G06N5/022"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121328689A/en"},{"publication_number":"KR20260007145A","title":"AI-based Smart Environmental Optimization","abstract":"본 발명은 사용자의 복수 생체 신호, 환경 요인, 그리고 생산 계획 및 설비 가동 상태 정보를 포함하는 외부 운영 데이터를 인공지능으로 통합 분석하여 조명, 냉난방, 공기질 및 기타 설비를 개인 맞춤형으로 통합 제어하는 인공지능 기반 생체 신호 연동 인간 중심 통합 환경 제어 시스템 및 방법에 관한 것이다. 특히, 본 발명은 AI의 예측 및 학습 기능을 통해 수면 유도, 스트레스 완화, 인지 능력 향상과 같은 의학적/생체학적 효과를 목표로 하면서도, 생산성 유지 또는 향상을 최우선으로 고려하여 환경을 최적화하며, 에너지 효율 극대화 및 지속적인 성능 고도화를 특징으로 한다. The present invention relates to an AI-based biosignal-linked human-centered integrated environmental control system and method that integrates and analyzes external operational data, including a user's multiple biosignals, environmental factors, and production plan and facility operation status information, using AI to provide personalized, integrated control of lighting, heating and cooling, air quality, and other facilities. In particular, the present invention utilizes AI's predictive and learning capabilities to target medical and biological benefits, such as sleep induction, stress relief, and cognitive enhancement, while prioritizing maintaining or enhancing productivity while optimizing the environment. Furthermore, the present invention features maximizing energy efficiency and continuously enhancing performance.","assignee":"구교선; 구현우","inventors":["구교선","구현우"],"publication_date":"2026-01-13","filing_date":"2025-12-18","priority_date":"2025-12-18","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","A","A61","A61B","A61B5/00","A61B5/01","A","A61","A61B","A61B5/00","A61B5/02","A61B5/024","A","A61","A61B","A61B5/00","A61B5/02","A61B5/024","A61B5/02405","A","A61","A61B","A61B5/00","A61B5/145","A61B5/14542","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/369","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4806","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260007145A/en"},{"publication_number":"CN121330195A","title":"A method and system for intelligent parsing and 3D modeling of road drawings","abstract":"The invention discloses an intelligent analysis and three-dimensional modeling method and system for road drawings, which comprise the steps of obtaining drawing data, carrying out content identification on the obtained drawing data, carrying out data screening and extraction based on the identified data, converting the data into structural data after cleaning and correction, carrying out secondary correction, and finally constructing a three-dimensional model according to the corrected data. According to the scheme, information extraction in the road drawing is carried out by utilizing a plurality of different classification or identification models before modeling, targeted design is carried out respectively on table data and image data, and data cleaning and correction are carried out successively after data extraction, so that the scheme can rapidly and accurately extract multi-dimensional road parameter information in the road drawing, lay a solid foundation for subsequent three-dimensional modeling, remarkably improve modeling precision in a complex drawing scene, and effectively solve the problem that a traditional method relies on an inefficient mode of manual transcription, trial calculation and repeated correction.","assignee":"Jiangsu Dinoni Information Technology Co ltd","inventors":["晁阳","吕泰达","杨嘉佳","陈瑜嘉","焦东","樊爽爽","彭林子","杨旻"],"publication_date":"2026-01-13","filing_date":"2025-12-18","priority_date":"2025-12-18","cpc_codes":["G","G06","G06T","G06T17/00","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N5/00","G06N5/02","G","G06","G06V","G06V10/00","G06V10/20","G06V10/26","G","G06","G06V","G06V10/00","G06V10/20","G06V10/28","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/75","G","G06","G06V","G06V10/00","G06V10/70","G06V10/762","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V30/00","G06V30/10","G06V30/19","G","G06","G06V","G06V30/00","G06V30/40","G06V30/41","G06V30/412","G","G06","G06V","G06V30/00","G06V30/40","G06V30/42","G06V30/422","G","G06","G06T","G06T2200/00","G06T2200/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121330195A/en"},{"publication_number":"KR20260005832A","title":"Electronic device and method for controlling thereof","abstract":"전자 장치 및 이의 제어 방법이 개시된다. 본 개시의 전자 장치는 제1 초점 거리를 갖는 제1 렌즈가 포함된 제1 카메라, 제1 초점 거리와 다른 제2 초점 거리를 갖는 제2 렌즈가 포함된 제2 카메라, 메모리 및 제1 카메라를 통해 전자 장치의 주변을 촬영하여 획득된 제1 영상에 포함된 제1 객체에 대한 정보를 획득하고, 획득된 제1 객체에 대한 정보에 기초하여 제1 객체와 연관된 제2 객체에 대한 정보가 존재한다고 식별되면, 제2 카메라를 통해 제2 영상을 획득하고, 획득된 제2 영상 상에 제2 객체가 포함되는지 여부에 기초하여, 제1 카메라를 이용하여 고속 촬영을 수행할 시점을 결정하는 프로세서를 포함할 수 있다. An electronic device and a method for controlling the same are disclosed. The electronic device of the present disclosure may include a first camera including a first lens having a first focal length, a second camera including a second lens having a second focal length different from the first focal length, a memory, and a processor configured to capture a periphery of the electronic device through the first camera to acquire information about a first object included in the acquired first image, and, if it is determined that information about a second object associated with the first object exists based on the acquired information about the first object, to acquire a second image through the second camera, and to determine a time point at which to perform high-speed photography using the first camera based on whether the second object is included in the acquired second image.","assignee":"삼성전자주식회사","inventors":["강희윤","강승수","최현수"],"publication_date":"2026-01-12","filing_date":"2025-12-19","priority_date":"2020-10-28","cpc_codes":["H","H04","H04N","H04N23/00","H04N23/60","H04N23/61","H04N23/611","G","G03","G03B","G03B15/00","G03B15/16","G","G03","G03B","G03B39/00","G","G06","G06N","G06N20/00","G","G06","G06N","G06N5/00","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06T","G06T7/00","G06T7/10","G06T7/13","G","G06","G06T","G06T7/00","G06T7/20","H","H04","H04N","H04N23/00","H04N23/45","H","H04","H04N","H04N23/00","H04N23/60","H","H04","H04N","H04N23/00","H04N23/60","H04N23/61","H","H04","H04N","H04N23/00","H04N23/60","H04N23/667","H","H04","H04N","H04N23/00","H04N23/90","H","H04","H04N","H04N23/00","H04N23/95","H","H04","H04N","H04N23/00","H04N23/95","H04N23/951","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260005832A/en"},{"publication_number":"KR20260005195A","title":"Method, apparatus and program for big data-based advertisement optimization","abstract":"본 발명의 다양한 실시예에 따른 빅데이터 기반 광고 최적화 방법이 개시된다. 상기 방법은: 빅데이터를 기초로 복수의 광고 매체 별 광고 효율을 예측하는 단계; 및 예측 광고 효율 및 광고 조건 정보를 기초로 광고 최적화 결과를 획득하는 단계;를 포함하고, 상기 광고 효율을 예측하는 단계는, 상기 빅데이터에서 특정 주차에 해당하는 과거 데이터를 기반으로 상기 특정 주차의 광고 효율을 예측하는 제1 모델을 통해 제1 광고 효율을 예측하는 단계; 상기 빅데이터에서 주차를 Binary 형태의 범주형 변수로 설정하여 상기 특정 주차의 광고 효율을 예측하는 제2 모델을 통해 제2 광고 효율을 예측하는 단계; 및 상기 제1 모델 및 상기 제2 모델 중 예측 정확도가 상대적으로 높은 어느 하나의 모델을 이용하여 상기 광고 효율을 예측하는 단계;를 포함할 수 있다. A big data-based advertising optimization method according to various embodiments of the present invention is disclosed. The method includes: a step of predicting advertising efficiency for each of a plurality of advertising media based on big data; and a step of obtaining an advertising optimization result based on the predicted advertising efficiency and advertising condition information; wherein the step of predicting advertising efficiency may include a step of predicting a first advertising efficiency through a first model that predicts the advertising efficiency of a specific parking lot based on past data corresponding to the specific parking lot in the big data; a step of predicting a second advertising efficiency through a second model that predicts the advertising efficiency of the specific parking lot by setting the parking lot in the big data as a categorical variable in the form of a binary number; and a step of predicting the advertising efficiency using any one model having a relatively high prediction accuracy among the first model and the second model.","assignee":"주식회사 임팩트에이아이","inventors":["박성혁","이민형"],"publication_date":"2026-01-09","filing_date":"2025-12-31","priority_date":"2024-01-24","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0242","G06Q30/0244","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q30/00","G06Q30/02","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0242","G06Q30/0246","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0272","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0273","G06Q30/0274"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260005195A/en"},{"publication_number":"KR20260005184A","title":"Apparatus and method for measuring weight of livestock based on image","abstract":"본 발명의 일 실시예에 따른 영상 기반 동물 성장 관리 장치는 대상체를 촬영하는 적어도 하나 이상의 카메라로부터 영상을 수신하는 통신 모듈; 동물 성장을 관리하는 프로그램이 저장된 메모리; 및 메모리에 저장된 프로그램을 실행하는 프로세서를 포함하며, 프로그램은, 제1 카메라로부터 수신한 영상에서 가축방으로 들어가는 전체 그룹의 제1 시점에서의 제1 무게 정보를 생성하고, 제2 카메라로부터 수신한 영상에서 가축방 내에 있는 전체 그룹의 제2 시점에서의 제2 무게 정보를 생성하고, 제1 카메라로부터 수신한 영상에서 가축방에서 나가는 부분 그룹의 제2 시점에서의 제2 무게 정보를 생성하고, 전체 그룹의 제2 무게 정보와 부분 그룹의 제2 무게 정보에 기초하여 가축방에 남아 있는 잔여 그룹의 제2 시점에서의 제2 무게 정보를 생성하고, 전체 그룹의 제2 무게 정보 중 평균 무게에 대한 부분 그룹과 잔여 그룹의 개체수 비율을 각각 계산하고, 계산된 각 개체수 비율을 전체 그룹의 제1 시점에서의 제1 무게 정보 중 평균 무게에 적용하여, 부분 그룹의 제1 시점에서의 제1 평균 무게와 잔여 그룹의 제1 시점에서의 제1 평균 무게를 역 계산함으로써, 부분 그룹의 제1 시점에서의 제1 무게 정보와 잔여 그룹의 제1 시점에서의 제1 무게 정보를 생성하되, 각 그룹 별 무게 정보는 개체수, 동물 개체 별 무게 리스트 및 평균 무게를 포함한다. An image-based animal growth management device according to one embodiment of the present invention comprises: a communication module for receiving an image from at least one camera that photographs a subject; a memory for storing a program for managing animal growth; And a processor executing a program stored in the memory, wherein the program generates first weight information at a first point in time of the entire group entering the livestock pen from an image received from a first camera, generates second weight information at a second point in time of the entire group within the livestock pen from an image received from a second camera, generates second weight information at a second point in time of the partial group exiting the livestock pen from an image received from the first camera, generates second weight information at a second point in time of the residual group remaining in the livestock pen based on the second weight information of the entire group and the second weight information of the partial group, calculates the population ratios of the partial group and the residual group with respect to the average weight among the second weight information of the entire group, and applies each calculated population ratio to the average weight among the first weight information of the entire group at the first point in time, thereby inversely calculating the first average weight at the first point in time of the partial group and the first average weight at the first point in time of the residual group, thereby generating the first weight information at the first point in time of the partial group and the first weight information at the first point in time of the residual group, wherein the weight information for each group includes the number of individuals, a list of individual animal weights, and an average weight.","assignee":"인트플로우 주식회사","inventors":["주소현","신동해","정진우","안형준","전광명"],"publication_date":"2026-01-09","filing_date":"2025-12-23","priority_date":"2024-06-04","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/02","A","A01","A01K","A01K29/00","A01K29/005","G","G01","G01G","G01G17/00","G01G17/08","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06T","G06T7/00","G06T7/60","G","G06","G06V","G06V10/00","G06V10/70","G06V10/766","G","G06","G06V","G06V40/00","G06V40/10"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260005184A/en"},{"publication_number":"KR20260005183A","title":"Video frame segmentation using reduced resolution neural network and masks from previous frames","abstract":"비디오 프레임 분할을 위한 예시적인 장치는 분할될 현재 비디오 프레임을 수신하는 수신기를 포함한다. 장치는 이전 프레임에 대응하는 분할 마스크를 포함하는 이전 마스크를 수신하고 이전 마스크 및 현재 비디오 프레임에 기초하여 현재 비디오 프레임에 대한 분할 마스크를 생성하는 분할 신경 네트워크도 포함한다. An exemplary device for video frame segmentation includes a receiver that receives a current video frame to be segmented. The device also includes a segmentation neural network that receives a previous mask including a segmentation mask corresponding to a previous frame and generates a segmentation mask for the current video frame based on the previous mask and the current video frame.","assignee":"인텔 코포레이션","inventors":["아미르 고렌","노암 엘론","노암 레비"],"publication_date":"2026-01-09","filing_date":"2025-12-23","priority_date":"2020-06-25","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/40","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06N","G06N3/00","G06N3/02","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06T","G06T3/00","G06T3/40","G06T3/4046","G","G06","G06T","G06T5/00","G06T5/70","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G06","G06T","G06T7/00","G06T7/10","G06T7/194","G","G06","G06T","G06T7/00","G06T7/50","G","G06","G06V","G06V10/00","G06V10/20","G06V10/26","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/40","G06V20/49","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G06V40/161","G06V40/165","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G06V40/168","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10016"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260005183A/en"},{"publication_number":"KR20260005176A","title":"Dynamic product recommendations on affiliate website","abstract":"동적으로 생성된 제품 추천을 제휴 웹사이트에 제공하기 위한 시스템 및 방법은 제품 데이터베이스의 각 제품에 관한 복수의 제품 특징을 생성하고, 제품 특징의 양과 관련된 길이를 가질 수 있는 벡터로 제품 데이터를 인코딩하고, 그리고 제품 참여 데이터에 기초하여 적어도 두 개의 제품 사이의 관련성 정도를 나타낼 수 있는 수치적 거리를 결정함으로써 복수의 제품의 각각의 제품과 관련된 제품 데이터를 임베딩하고, 제1 기계 학습 모델을 사용하여 제품 참여 데이터에 기초하여 후보 제품을 결정하고, 제2 기계 학습 모델을 사용하여 적어도 임베딩된 제품 데이터 및 제휴사 참여 데이터에 기초하여 후보 제품의 셀렉션을 결정하고, 그리고 식별할 수 없는 사용자 디바이스 상에 표시하기 위해 후보 제품의 셀렉션을 제휴 서버에 제공하는 것을 포함할 수 있다. A system and method for providing dynamically generated product recommendations to an affiliate website may include generating a plurality of product features for each product in a product database, encoding the product data into a vector, the vector having a length that may be related to a quantity of the product features, and embedding product data associated with each of the plurality of products by determining a numerical distance that may indicate a degree of relevance between at least two products based on product engagement data, determining candidate products based on the product engagement data using a first machine learning model, determining a selection of candidate products based on at least the embedded product data and the affiliate engagement data using a second machine learning model, and providing the selection of candidate products to an affiliate server for display on an unidentifiable user device.","assignee":"쿠팡 주식회사","inventors":["정도윤","신리 바오","수멍 람"],"publication_date":"2026-01-09","filing_date":"2025-12-22","priority_date":"2022-06-16","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0631","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2135","G06F18/21355","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0207","G06Q30/0222","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0207","G06Q30/0239","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0251","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0251","G06Q30/0253","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0251","G06Q30/0255","G06Q30/0256","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0282","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0603","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0633","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0641","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0641","G06Q30/0643"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260005176A/en"},{"publication_number":"KR20260005169A","title":"Optical measurement method and optical metrology device removing undesired spectral effects","abstract":"미지의 구조에 광학적으로 결합된 관심 구조(SOI)를 포함하는 샘플의 광학 측정은 생성된 스펙트럼 신호로부터 SOI와 연관된 키 파라미터와 상관되는 스펙트럼 변동을 추출하고, 키 파라미터에 무관한 미지의 구조로부터 스펙트럼 변동을 제거함으로써 광학적으로 측정된다. 오프라인 프로세스를 사용하여 키 파라미터에 무관한 스펙트럼 신호로부터 스펙트럼 변동을 제거한 후에 재구축된 스펙트럼 신호를 사용하여 다수의 교정 측정으로부터 물리 기반 모델을 생성한다. 기계 학습 모델은 SOI와 연관된 키 파라미터에 상관된 적어도 스펙트럼 변동을 사용하여 추가로 생성될 수 있다. 인-라인 프로세스에서, 샘플로부터 스펙트럼 신호를 필터링하여 미지의 구조로부터 스펙트럼 효과를 제거하고 물리 기반 모델을 사용하거나 학습된 기계 학습 모델을 사용함으로써 샘플을 측정한다. Optical measurements of a sample containing a structure of interest (SOI) optically coupled to an unknown structure are performed by extracting spectral fluctuations correlated with key parameters associated with the SOI from the generated spectral signal and removing spectral fluctuations from the unknown structure that are unrelated to the key parameters. After removing spectral fluctuations from the spectral signal that are unrelated to the key parameters using an offline process, a physics-based model is generated from multiple calibration measurements using the reconstructed spectral signal. A machine learning model can be further generated using at least the spectral fluctuations correlated to key parameters associated with the SOI. In an in-line process, the spectral signal from the sample is filtered to remove spectral effects from the unknown structure and the sample is measured using the physics-based model or the trained machine learning model.","assignee":"온투 이노베이션 아이엔씨.","inventors":["페타르 주벨라","징쉥 쉬","웨이 밍 치에우","제 리"],"publication_date":"2026-01-09","filing_date":"2025-12-19","priority_date":"2022-07-08","cpc_codes":["G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/95","G01N21/9501","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/95","G01N21/956","G","G01","G01N","G01N21/00","G01N21/17","G01N21/25","G","G01","G01J","G01J3/00","G01J3/02","G01J3/0262","G","G01","G01J","G01J3/00","G01J3/02","G01J3/0297","G","G01","G01J","G01J3/00","G01J3/28","G","G01","G01N","G01N21/00","G01N21/01","G","G01","G01N","G01N21/00","G01N21/17","G01N21/21","G01N21/211","G","G01","G01N","G01N21/00","G01N21/17","G01N21/25","G01N21/31","G","G01","G01N","G01N21/00","G01N21/17","G01N21/25","G01N21/31","G01N21/35","G","G01","G01N","G01N21/00","G01N21/17","G01N21/41","G01N21/45","G","G01","G01N","G01N21/00","G01N21/17","G01N21/55","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06N","G06N20/00","G","G01","G01N","G01N21/00","G01N21/17","G01N21/21","G01N21/211","G01N2021/213","G","G01","G01N","G01N21/00","G01N21/17","G01N21/25","G01N21/31","G01N21/35","G01N2021/3595","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G01N2021/8883","G","G01","G01N","G01N2201/00","G01N2201/12","G01N2201/129","G01N2201/1296","H","H10","H10P","H10P74/00","H10P74/20","H10P74/203"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260005169A/en"},{"publication_number":"KR20260005154A","title":"Investment consultation chatbot service providing system and method performing thereof","abstract":"본 발명의 일실시예를 따르는 소셜 트레이딩 서비스 제공 서버는, 복수의 자산 거래 플랫폼 서버로부터 자산 딜러 별 거래 정보를 수신하여 딜러 정보 데이터베이스에 저장하는 딜러 정보 수집부; 딜러의 거래 정보 중 딜러 액션 정보를 기초로 매입 시점부터 매각 시점까지의 기간을 확인하고 상기 기간의 범위에 따라 자산 딜러의 투자 기간 그룹을 결정하고, 투자 기간 그룹 별로 수익률을 결정하고, 투자 기간 그룹 별로 그룹핑된 자산 딜러를 수익률에 따라 다시 한번 소그룹으로 분류하는 딜러 분류부; 투자 기간 그룹 별로 해당 그룹의 소분류 그룹으로 분류된 자산 딜러가 투자한 자산 종류를 추출하고, 상기 자산 딜러가 투자한 자산의 매입 시점부터 매각 시점 사이에서, 상기 자산의 매입 가격보다 높은 제 1 매도 가능 가격인 제 1 매도 가능 시점을 결정하고, 상기 제1 매도 가능 가격보다 낮지만 상기 매입 가격보다 높은 제 2 매도 가능 가격인 제 2 매도 가능 시점을 결정하고, 상기 제 1 매도 가능 가격에 상기 제 1 매도 가능 시점의 시장 상황에 해당하는 가중치를 적용한 후 평균화하여 제 1 평균 가격을 산출하고, 상기 제 2 매도 가능 가격에 상기 제 2 매도 가능 시점의 시장 상황에 해당하는 가중치를 적용한 후 평균화하여 제 2 평균 가격을 산출하고, 산기 제 1 평균 가격 및 상기 제 2 평균 가격의 차이 및 수익 금액을 비교하여 수익 금액이 높아지는 시점을 딜러 매도 시점으로 예측하는 학습 모델을 생성하는 딜러 별 학습 모델 생성부를 포함하는 것을 특징으로 한다. A social trading service providing server according to one embodiment of the present invention comprises: a dealer information collection unit that receives transaction information for each asset dealer from a plurality of asset trading platform servers and stores the transaction information in a dealer information database; a dealer classification unit that verifies a period from a purchase time to a sale time based on dealer action information among the dealer's transaction information, determines an investment period group of the asset dealer according to the range of the period, determines a rate of return for each investment period group, and further classifies the asset dealers grouped by investment period group into smaller groups according to the rate of return; The method is characterized by including a dealer-specific learning model generation unit that extracts the types of assets invested by asset dealers classified into subgroups of each group for each investment period group, determines a first saleable point in time, which is a first saleable price higher than the purchase price of the asset between the purchase time and the sale time of the asset invested by the asset dealer, determines a second saleable point in time, which is a second saleable price lower than the first saleable price but higher than the purchase price, calculates a first average price by applying a weight corresponding to the market situation at the first saleable point in time to the first saleable price and then averaging the results, and calculates a second average price by applying a weight corresponding to the market situation at the second saleable point in time to the second saleable price and then averaging the results, and generates a learning model that compares the difference between the first average price and the second average price and the profit amount to predict a point in time when the profit amount increases as the dealer sale point.","assignee":"주식회사 아데나소프트웨어","inventors":["정승우"],"publication_date":"2026-01-09","filing_date":"2025-12-17","priority_date":"2023-07-31","cpc_codes":["G","G06","G06Q","G06Q40/00","G06Q40/04","G06Q40/042","G06Q40/0421","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0281","G","G06","G06Q","G06Q40/00","G06Q40/06","G06Q40/063","G06Q40/0631","G","G06","G06Q","G06Q50/00","G06Q50/50","H","H04","H04L","H04L51/00","H04L51/02"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260005154A/en"},{"publication_number":"CN121302567A","title":"TPT-based front subframe structure MaOP method, equipment, media, and program products","abstract":"本发明公开了基于TPT的前副车架结构MaOP方法、设备、介质和程序产品，包括：（1）基于前副车架纵梁及载荷分析，构建能同时优化重量、模态、强度与刚度的MaOP设计模型；（2）基于多样性准则与拉丁超立方产生精英种群，获得精英种群的各目标值并建立数据集，构建径向基函数模型；（3）设计TPT驱动的协同进化操作产生候选车架集合；（4）基于优势、劣势目标集构造个体潜力评估准则，筛选最优候选车架；（5）获取最优候选车架的各目标值，更新数据集及径向基函数模型，返回步骤（3），直至所有优化目标达到要求，输出最优参数取值。本发明采用TPT驱动的协同进化机制，结合径向基函数模型预测，能够有效平衡针对前副车架的多性能指标寻优进程。 This invention discloses a TPT-driven MaOP method, device, medium, and program product for front subframe structures, including: (1) constructing a MaOP design model that can simultaneously optimize weight, modality, strength, and stiffness based on the analysis of the front subframe longitudinal beams and loads; (2) generating an elite population based on diversity criteria and Latin hypercube, obtaining the target values of the elite population and establishing a dataset, and constructing a radial basis function model; (3) designing a TPT-driven co-evolutionary operation to generate a candidate frame set; (4) constructing an individual potential evaluation criterion based on the set of advantages and disadvantages, and screening the optimal candidate frame; (5) obtaining the target values of the optimal candidate frame, updating the dataset and radial basis function model, returning to step (3), until all optimization targets meet the requirements, and outputting the optimal parameter values. This invention adopts a TPT-driven co-evolutionary mechanism, combined with radial basis function model prediction, which can effectively balance the optimization process of multiple performance indicators for the front subframe.","assignee":"Nanchang University","inventors":["杨赞","褚福齐","杜兴","吴嘉欣","张旭","刘信风","章嘉乐","谭祎康"],"publication_date":"2026-01-09","filing_date":"2025-12-15","priority_date":"2025-12-15","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/10","G06F30/15","G","G06","G06F","G06F30/00","G06F30/20","G06F30/23","G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06F","G06F2111/00","G06F2111/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121302567A/en"},{"publication_number":"KR20260004279A","title":"Method and system for providing recommendation information related to photography","abstract":"제 1 카메라를 통해 인식되는 프리뷰 이미지에 포함된 피사체를 식별하는 단계; 식별된 피사체 주변의 빛과 관련된 정보를 획득하는 단계; 식별된 피사체의 정보 및 피사체 주변의 빛과 관련된 정보를 이용하여, 추천 촬영 구도를 결정하는 단계; 및 결정된 추천 촬영 구도에 관한 정보를 제공하는 단계를 포함하는 전자 장치가 사진 촬영과 관련된 추천 정보를 제공하는 방법이 개시된다. A method for providing recommended information related to photographing is disclosed, wherein an electronic device includes a step of identifying a subject included in a preview image recognized through a first camera; a step of obtaining information related to light surrounding the identified subject; a step of determining a recommended photographing composition using information about the identified subject and information related to light surrounding the subject; and a step of providing information about the determined recommended photographing composition.","assignee":"삼성전자주식회사","inventors":["정재호","성열탁"],"publication_date":"2026-01-08","filing_date":"2025-12-18","priority_date":"2017-12-01","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","H","H04","H04N","H04N23/00","H04N23/60","H04N23/64","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","H","H04","H04M","H04M1/00","H04M1/72","H04M1/724","H04M1/72403","H","H04","H04N","H04N1/00","H04N1/00127","H04N1/00204","H04N1/00244","H","H04","H04N","H04N23/00","H04N23/45","H","H04","H04N","H04N23/00","H04N23/60","H04N23/61","H","H04","H04N","H04N23/00","H04N23/60","H04N23/62","H","H04","H04N","H04N23/00","H04N23/60","H04N23/63","H04N23/631","H04N23/632","H","H04","H04N","H04N23/00","H04N23/60","H04N23/63","H04N23/633","H","H04","H04N","H04N23/00","H04N23/60","H04N23/63","H04N23/633","H04N23/635","H","H04","H04N","H04N23/00","H04N23/60","H04N23/695","H","H04","H04N","H04N23/00","H04N23/70","H04N23/71","H","H04","H04N","H04N23/00","H04N23/90","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","H","H04","H04N","H04N2201/00","H04N2201/0077","H04N2201/0084"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260004279A/en"},{"publication_number":"KR20260004264A","title":"Apparatus and method of anomaly detection using neural network","abstract":"뉴럴 네트워크를 이용한 이상 검출(anomaly detection) 장치 및 방법이 개시된다. 일 실시예에 따른 이상 검출(anomaly detection) 장치는, 데이터를 수신하는 수신기, 및 정상 데이터에 대해 미리 설정된 그룹과 상기 데이터 사이의 거리에 기초하여 상기 데이터에 대응하는 복수의 특징을 추출하고, 상기 복수의 특징에 기초하여 상기 복수의 그룹에 대응하는 복수의 OOD(Out Of Distribution) 점수를 계산하고, 상기 복수의 OOD 점수에 기초하여 상기 데이터의 정상 여부를 검출하는 프로세서를 포함할 수 있다. An anomaly detection device and method using a neural network are disclosed. An anomaly detection device according to one embodiment may include a receiver that receives data, and a processor that extracts a plurality of features corresponding to the data based on a distance between a preset group for normal data and the data, calculates a plurality of OOD (Out Of Distribution) scores corresponding to the plurality of groups based on the plurality of features, and detects whether the data is normal based on the plurality of OOD scores.","assignee":"주식회사 엘로이랩","inventors":["유광선","김명환"],"publication_date":"2026-01-08","filing_date":"2025-12-15","priority_date":"2022-05-16","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260004264A/en"},{"publication_number":"KR20260003622A","title":"Cooking device and operating method thereof","abstract":"본 개시의 실시 예에 따른 조리 기기는 디스플레이부; 조리실; 상기 조리실을 가열하는 가열부; 상기 조리실의 내부에 위치한 음식을 촬영하는 카메라; 및 이미지의 속성의 수치를 조작하여 변화된 이미지를 생성하는 이미지 생성 모델을 이용하여, 상기 카메라를 통해 촬영된 상기 음식의 이미지로부터 상기 음식의 조리 진행 단계를 나타내는 복수의 예상 이미지들을 생성하고, 생성된 복수의 예상 이미지들을 상기 디스플레이 상에 표시하고, 상기 복수의 예상 이미지들 중 어느 하나를 선택하는 명령에 따라 선택된 예상 이미지에 매칭되는 조리 정보를 이용하여 상기 음식을 조리하도록 상기 가열부를 제어하는 프로세서를 포함할 수 있다. A cooking appliance according to an embodiment of the present disclosure may include a display unit; a cooking chamber; a heating unit for heating the cooking chamber; a camera for photographing food located inside the cooking chamber; and a processor for generating a plurality of predicted images representing cooking progress stages of the food from an image of the food photographed through the camera using an image generation model that generates a changed image by manipulating numerical values of attributes of the image, and for controlling the heating unit to display the generated plurality of predicted images on the display and cook the food using cooking information matching a predicted image selected in response to a command for selecting any one of the plurality of predicted images.","assignee":"엘지전자 주식회사","inventors":["유은경","도호석"],"publication_date":"2026-01-07","filing_date":"2025-12-12","priority_date":"2023-05-11","cpc_codes":["F","F24","F24C","F24C7/00","F24C7/08","F24C7/082","F24C7/085","A","A47","A47J","A47J36/00","A47J36/32","A47J36/321","F","F24","F24C","F24C7/00","F24C7/08","F24C7/082","F24C7/086","G","G01","G01J","G01J1/00","G01J1/02","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T11/00","G","G06","G06T","G06T11/00","G06T11/10","G","G06","G06V","G06V20/00","G06V20/60","G06V20/68","H","H04","H04N","H04N23/00","H04N23/60","H04N23/61","H","H04","H04N","H04N7/00","H04N7/18","H04N7/183","G","G06","G06T","G06T2200/00","G06T2200/24"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260003622A/en"},{"publication_number":"KR20260002509A","title":"Method for providing autopilot public transportation service and device for the same","abstract":"과거 날짜에서 발생했던 수요콜들의 발생이력들을 기초로 특정 시구간에서 상기 자율주행 대중교통 서비스를 이용하는 승객들 각각에 대하여 제공된 상기 자율주행 대중교통 서비스의 제공시간의 통계값을 상기 자율주행 대중교통 서비스에 투입되는 자율주행 차량들의 투입 대수마다 시뮬레이션 하여 산출하고, 상기 산출된 통계값이 소정의 조건 을 만족시키는 상기 자율주행 차량들의 투입 대수들 중에서 최소 대수를 산출하고, 상기 산출된 최소 대수를 상기 현재 날짜에서의 상기 특정 시구간에서 상기 자유주행 대중교통 서비스를 위해 배차될 자율주행 차량들의 투입 대수로 결정하는 기술을 공개한다. A technology is disclosed that calculates statistical values of the provision time of the autonomous public transportation service provided to each passenger using the autonomous public transportation service in a specific time period based on the occurrence history of demand calls that occurred in the past date by simulating the number of autonomous vehicles deployed in the autonomous public transportation service, calculates the minimum number of autonomous vehicles deployed among the numbers of autonomous vehicles that satisfy a predetermined condition based on the generated statistical values, and determines the calculated minimum number as the number of autonomous vehicles to be deployed for the free-driving public transportation service in the specific time period on the current date.","assignee":"국립한국교통대학교산학협력단","inventors":["김현"],"publication_date":"2026-01-06","filing_date":"2025-12-18","priority_date":"2022-11-18","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/40","G","G01","G01C","G01C21/00","G01C21/26","G01C21/34","G01C21/3446","G","G06","G06F","G06F17/00","G06F17/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260002509A/en"},{"publication_number":"KR20260002503A","title":"Method, computer device, and computer program to provide marketing message as benefit information","abstract":"마케팅 메시지를 혜택 정보로 제공하기 위한 방법, 컴퓨터 장치, 및 컴퓨터 프로그램이 개시된다. 마케팅 메시지 제공 방법은, 사업자에 의해 최근 일정 시간 동안 발송된 마케팅 메시지 중 일정 조건에 부합하는 메시지를 선별하는 단계; 및 상기 일정 조건에 부합하는 메시지를 혜택 피드로 구성하여 노출하는 단계를 포함할 수 있다. A method, a computer device, and a computer program for providing marketing messages as benefit information are disclosed. The method for providing marketing messages may include the steps of selecting messages that meet certain conditions from among marketing messages sent by a business operator over a recent period of time; and the steps of organizing and displaying messages that meet the certain conditions as a benefit feed.","assignee":"네이버 주식회사","inventors":["서유경","구송이","김세철","이수연","임소진","안가영","오혜림","윤경희","이소명","유선우","박찬혜","김성민","임영규","김준모","임동섭","안창규","이경환","유현도","손현준"],"publication_date":"2026-01-06","filing_date":"2025-12-17","priority_date":"2022-06-30","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0277","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/955","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0251","G06Q30/0254","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0251","G06Q30/0267","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0251","G06Q30/0269","G06Q30/0271","G","G06","G06Q","G06Q50/00","G06Q50/50","H","H04","H04L","H04L51/00","H04L51/04"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260002503A/en"},{"publication_number":"KR20260002488A","title":"System, method and apparatus for multi-agent reinforcement learning","abstract":"본 개시의 일 실시예에 따른 다중 에이전트 강화학습(Multi-Agent Reinforcement Learning) 시스템은, 제1 에이전트 및 제2 에이전트를 포함하는 다중 에이전트; 다중 에이전트의 각 에이전트마다 마련되는 히스토리 인코더; 하나 이상의 명령어들을 저장하는 메모리; 및 상기 메모리에 저장된 상기 하나 이상의 명령어들을 실행하는 적어도 하나의 프로세서를 포함하며, 상기 적어도 하나의 프로세서는 상기 하나 이상의 명령어들을 실행함으로써, 상기 제1 에이전트의 관측 데이터를 제1 히스토리 인코더에 입력하여 상기 제1 에이전트의 제1 히스토리 정보를 생성하고, 상기 제2 에이전트의 관측 데이터를 제2 히스토리 인코더에 입력하여 상기 제2 에이전트의 제2 히스토리 정보를 생성할 수 있다. A multi-agent reinforcement learning system according to one embodiment of the present disclosure comprises: multiple agents including a first agent and a second agent; a history encoder provided for each agent of the multiple agents; a memory storing one or more commands; and at least one processor executing the one or more commands stored in the memory, wherein the at least one processor can input observation data of the first agent into the first history encoder to generate first history information of the first agent by executing the one or more commands, and input observation data of the second agent into the second history encoder to generate second history information of the second agent.","assignee":"주식회사 Lg 경영개발원","inventors":["임우형","이강훈","윤든솔","홍성훈","정휘영"],"publication_date":"2026-01-06","filing_date":"2025-12-16","priority_date":"2024-03-11","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260002488A/en"},{"publication_number":"KR20260002465A","title":"Ai based job matching method and apparatus","abstract":"본 발명에 따른 일자리 매칭 방법은 근로자로부터 일자리 매칭정보를 입력받는 입력 단계, 상기 일자리 매칭정보에 기초하여 하나 이상의 근로현장을 추출하는 추출 단계, 상기 하나 이상의 근로현장의 기상정보를 획득하는 획득 단계, 기상정보에 기초하여 상기 하나 이상의 근로현장에 대한 상기 근로자의 출석확률을 예측하는 예측 단계, 및 상기 출석확률에 기초하여 상기 근로자에게 매칭할 근로현장 및 근로일자를 결정하는 결정 단계를 포함한다. 본 발명에 따른 일자리 매칭 방법 및 장치에 의하면, 날씨, 기후 등의 환경 조건을 고려한 출석률 예측모델을 이용하기 때문에 근로자의 출석율을 향상시키고, 구인업체에 대한 근로자 확보를 보장할 수 있다. 또한, 근로자의 기술과 경험에 따라 일자리 매칭이 이루어지므로, 근로자는 보다 적절한 일자리를 매칭받아 커리어를 쌓을 수 있다. 이에 따라, 건설업계의 인력부족 문제를 해결하고, 근로자와 구인업체 모두에게 혜택을 제공할 수 있다. A job matching method according to the present invention comprises an input step of receiving job matching information from a worker, an extraction step of extracting one or more work sites based on the job matching information, an acquisition step of obtaining weather information for the one or more work sites, a prediction step of predicting the worker's attendance probability for the one or more work sites based on the weather information, and a determination step of determining a work site and work date to be matched with the worker based on the attendance probability. The job matching method and device according to the present invention utilize an attendance rate prediction model that takes into account environmental conditions such as weather and climate, thereby improving worker attendance rates and ensuring worker retention for recruiting companies. Furthermore, since job matching is performed based on the worker's skills and experience, the worker is matched with more appropriate jobs and can build his or her career. Accordingly, the labor shortage problem in the construction industry can be resolved, providing benefits to both workers and recruiting companies.","assignee":"김진호","inventors":["김진호","김성호","김덕호","성현우"],"publication_date":"2026-01-06","filing_date":"2025-12-15","priority_date":"2023-05-09","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/105","G06Q10/1053","G","G01","G01W","G01W1/00","G01W1/02","G","G01","G01W","G01W1/00","G01W1/10","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06311","G06Q10/063118","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06393","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06398","G","G07","G07C","G07C1/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20260002465A/en"},{"publication_number":"CN121279757A","title":"Unmanned aerial vehicle-based transmission line inspection visualization early warning system and method","abstract":"本发明公开了基于无人机的输电线路巡检可视化预警系统及方法，涉及输电线路巡检技术领域，本发明通过引入多源关联深度编码机制，打破了传统巡检分析中静态设备属性与动态环境信息分离的局限，实现了对输电设备在特定环境上下文中的动态脆弱性评估。所述深度关联编码模块能够精确量化外部环境因素如强风、高温或高湿对不同类型、不同服役年限的输电设备的影响权重，从而输出一个高维度的设备‑环境交互特征向量，使风险评估具备高度的上下文依赖性。解决了现有技术中对所有设备采用无差别统一分析，造成资源浪费和风险遗漏的问题。 This invention discloses a visualization and early warning system and method for power transmission line inspection based on unmanned aerial vehicles (UAVs), belonging to the field of power transmission line inspection technology. By introducing a multi-source correlation deep coding mechanism, this invention breaks through the limitation of separating static equipment attributes from dynamic environmental information in traditional inspection analysis, enabling dynamic vulnerability assessment of power transmission equipment in specific environmental contexts. The deep correlation coding module can accurately quantify the impact weights of external environmental factors such as strong winds, high temperatures, or high humidity on different types and service lives of power transmission equipment, thereby outputting a high-dimensional equipment-environment interaction feature vector, making risk assessment highly context-dependent. This solves the problem of resource waste and risk omission caused by the undifferentiated uniform analysis of all equipment in existing technologies.","assignee":"Liaoning Power Energy Development Group Co ltd","inventors":["金成明","同东辉","姚振先","孙熙","孙建航","闫圣夫"],"publication_date":"2026-01-06","filing_date":"2025-12-11","priority_date":"2025-12-11","cpc_codes":["H","H02","H02G","H02G1/00","H02G1/02","G","G06","G06F","G06F18/00","G06F18/20","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06312","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06316","G","G06","G06Q","G06Q10/00","G06Q10/20","G","G06","G06Q","G06Q50/00","G06Q50/06","G","G08","G08B","G08B31/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121279757A/en"},{"publication_number":"CN121279759A","title":"Flexible job shop production scheduling and machine maintenance combined optimization method and system","abstract":"The invention belongs to the technical field of workshop scheduling and maintenance combination, and particularly relates to a flexible job workshop production scheduling and machine maintenance combination optimization method and system. The method comprises the steps of S1, obtaining basic information of joint optimization of production scheduling and machine maintenance of a flexible job shop, S2, constructing a joint optimization model of the production scheduling and the machine maintenance of the flexible job shop by taking the total cost of a minimum production system as an objective function, S3, converting the joint optimization model of the production scheduling and the machine maintenance of the flexible job shop into a Markov decision process, respectively defining states, actions, state transitions and rewards of the production scheduling and the machine maintenance, S4, designing a joint optimization method based on a digital twin and double-layer reinforcement learning algorithm based on the Markov decision process, and obtaining trained agents through training, and S5, applying the trained agents to practical joint optimization problems of the production scheduling and the machine maintenance of the flexible job shop.","assignee":"Hangzhou Dianzi University","inventors":["李晓","周培旭","阮渊鹏"],"publication_date":"2026-01-06","filing_date":"2025-12-11","priority_date":"2025-12-11","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06312","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/20","G","G06","G06Q","G06Q50/00","G06Q50/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121279759A/en"},{"publication_number":"CN121278361A","title":"PyraMamba-based multi-mode track prediction method","abstract":"A PyraMamba-based multi-mode track prediction method is realized through a data preprocessing module, a time sequence feature extraction module, a space feature extraction module and a Laplacian decoder, wherein original data is divided into a training set and a verification set after being processed through the data preprocessing module. In the training stage, a sequential feature extraction module builds PyraMamba a model based on a pyramid structure to extract sequential features of a track sequence, a spatial feature extraction module processes map vectors and the track sequence to model a spatial interaction relation between map elements and the track features, a Laplace decoder fuses space-time information to predict Laplace distribution parameters and corresponding modal probability of a plurality of candidate tracks in the future, and multi-modal prediction of the future tracks of traffic participants is achieved. Finally, the prediction precision and the robust performance of the verification model are evaluated through the verification set.","assignee":"Changchun University of Technology","inventors":["李绍松","周苏豫","黄俊凯","田丽媛","黄熙哲","王德鑫","卢晓晖","崔高健","施宏达"],"publication_date":"2026-01-06","filing_date":"2025-12-11","priority_date":"2025-12-11","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06F","G06F18/00","G06F18/20","G06F18/27","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121278361A/en"},{"publication_number":"CN121280638A","title":"Near-infrared auxiliary low light scene three-dimensional reconstruction method based on 3D Gaussian splatter","abstract":"The invention relates to the technical field of three-dimensional reconstruction, in particular to a near-infrared auxiliary low-light scene three-dimensional reconstruction method based on 3D Gaussian splats, which utilizes near-infrared images with high signal to noise ratio to lead Gaussian ellipsoids to generate so as to provide a stable geometric foundation for a visible light information missing region, respectively generates near-infrared rendering images and normal visible light rendering images through the sharing of geometric parameters of the Gaussian ellipsoids and respective opacity and color attributes of two modes in a rendering stage, forces the normal visible light images and the near-infrared images to be rendered to be aligned structurally through the loss of cross-mode structural similarity, and strictly restricts the recovery process of the colors of the visible light images by clear near-infrared structural information so as to ensure the accuracy and the authenticity of recovery results.","assignee":"Zhejiang University ZJU","inventors":["李樵风","杨承昀","章毅"],"publication_date":"2026-01-06","filing_date":"2025-12-10","priority_date":"2025-12-10","cpc_codes":["G","G06","G06T","G06T17/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","Y","Y02","Y02T","Y02T10/00","Y02T10/10","Y02T10/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121280638A/en"},{"publication_number":"CN121259512A","title":"Real-time monitoring method and system for few samples of abnormal behaviors of park","abstract":"The invention provides a real-time monitoring method and a system for few samples of abnormal behaviors of a park, which relate to the technical field of computers, and the method comprises the steps of acquiring a real-time monitoring video stream of the park as a query set, and calling a plurality of video samples of known categories from a preset abnormal behavior database as a support set; the method comprises the steps of processing video frame sequences in a query set and a support set respectively through a feature extraction network, extracting spatial features representing static details, time sequence features representing global evolution and action features representing inter-frame change intensity in parallel, carrying out weighting enhancement on the spatial features based on the motion intensity represented by the action features, fusing the enhanced spatial features with the time sequence features to generate comprehensive features, inputting the comprehensive features of the query set into a matching model to serve as query items, carrying out self-adaptive re-weighting on the comprehensive features of the support set, and judging whether abnormal behaviors exist in a real-time monitoring video stream of a park based on the query items and the re-weighted comprehensive features of the support set.","assignee":"China Post Intelligence Xi'an Technology Co ltd; China Posts And Telecommunications Equipment Group Co ltd; Sun Yat Sen University","inventors":["李婧","周凡","张扬眉","林格","王子寒","苏卓"],"publication_date":"2026-01-02","filing_date":"2025-12-08","priority_date":"2025-12-08","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/761","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/40","G06V20/41","G","G06","G06V","G06V20/00","G06V20/40","G06V20/46"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121259512A/en"},{"publication_number":"CN121258450A","title":"On-chain simulation system and method for construction project contract performance risk","abstract":"The embodiment of the application provides a simulation system and a simulation method on a construction project contract performance risk chain, which are applied to the technical field of project management and risk management, wherein the method comprises the steps of constructing an event state response table containing shield segment supply, tunnel tunneling quality and geological condition change risk events based on a construction contract; the method comprises the steps of establishing a multi-level state sequence comprising an initial state, a supply risk state, a quality risk state and a comprehensive risk state, setting risk weight and associated fund guarantee strategy for each state transition, deploying a main contract, an on-chain event processing engine and a verification sub-contract system on a alliance blockchain, establishing a guarantee fund pool, verifying construction data through the verification sub-contract, identifying a risk mode by using the on-chain event processing engine and generating a state transition instruction, and calculating fund adjustment quantity according to the instruction and the risk weight by the main contract. The method and the system realize full-flow automatic management of contract performance risks in subway tunnel engineering construction.","assignee":"Shanghai Baitong Xiangguan Technology Co ltd","inventors":["马小燕","施锦刚","李爽","王英","廖安安","欧阳哲","罗志亮"],"publication_date":"2026-01-02","filing_date":"2025-12-08","priority_date":"2025-12-08","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/103","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2431","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06Q","G06Q40/00","G06Q40/03","G","G06","G06Q","G06Q50/00","G06Q50/08","Y","Y02","Y02P","Y02P90/00","Y02P90/30"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121258450A/en"},{"publication_number":"CN121256655A","title":"Logging data abnormal point screening and correcting method and device based on hole elasticity theory, electronic equipment and storage medium","abstract":"The invention relates to the technical field of well logging data screening and correction, in particular to a well logging data abnormal point screening and correction method, device, electronic equipment and storage medium based on a hole elasticity theory, which comprises the steps of deleting data points which do not accord with a physical rationality verification rule in each curve; and obtaining a corresponding predicted curve of the lower part of the well diameter median by using a curve prediction model, obtaining a corresponding correction curve by weighted superposition, and correcting Kong Danli parameter curves to intensively represent the rest curves influenced by the well diameter abnormality based on the corrected curve. The method disclosed by the invention is used for merging multiple outlier score retrieval methods to identify independent outliers, so that independent outliers are accurately excavated, error identification of the independent outliers is avoided, each curve prediction model is constructed and used for fitting prediction of corresponding curves, and therefore, correlation outliers are accurately identified on the basis of internal correlation among fully excavated data.","assignee":"China National Petroleum Corp; CNPC Xibu Drilling Engineering Co Ltd","inventors":["毛仕卓","李哲","田泽华","卢雨潇","解俊昱","刘媛"],"publication_date":"2026-01-02","filing_date":"2025-12-08","priority_date":"2025-12-08","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2433","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2455","G06F16/24564","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2458","G06F16/2462","G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2135","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","Y","Y02","Y02A","Y02A90/00","Y02A90/30"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121256655A/en"},{"publication_number":"CN121261302A","title":"Emergency protection method for intelligent medium-voltage ring network switch cabinet","abstract":"The invention belongs to the technical field of intelligent power grid protection and discloses an emergency protection method of an intelligent medium-voltage ring network switch cabinet, which comprises the steps of obtaining dynamic electrical association diagrams by acquiring real-time electrical state data of each switch cabinet in ring network topology and carrying out multi-parameter electrical coupling analysis on the real-time electrical state data; performing fault identification based on the dynamic electrical correlation diagram to obtain a fault source node, and constructing a fault propagation path prediction matrix by using the position of the fault source node and the electrical coupling strength distribution characteristics; the method and the device for achieving the ring network power supply protection and the intelligent operation and maintenance of the ring network power supply have the advantages that the accuracy of fault isolation and the coordination of protection actions are remarkably improved, and high-quality technical support is provided for ring network power supply protection and intelligent operation and maintenance.","assignee":"Dongguan Nabaichuan Electronic Technological Co ltd","inventors":["李德高"],"publication_date":"2026-01-02","filing_date":"2025-12-08","priority_date":"2025-12-08","cpc_codes":["H","H02","H02H","H02H7/00","H02H7/22","H02H7/222","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2431","G","G06","G06F","G06F18/00","G06F18/20","G06F18/26","G","G06","G06N","G06N5/00","G06N5/01","H","H02","H02H","H02H1/00","H02H1/0092"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121261302A/en"},{"publication_number":"CN121260484A","title":"Index combination, method and system for predicting obesity-related metabolic syndrome","abstract":"The application relates to an index combination, a method and a system for predicting obesity-related metabolic syndrome, which comprises an index combination for predicting obesity-related metabolic syndrome, wherein the index combination comprises indexes including a low mitochondrial membrane potential percentage (NKG 2C + NK.MMP low %）、NKG2D + NKT cell) in NKG2C + NK cells (NKG 2D + NKT.MMP low %）、NKG2C + NKT cell) and a low mitochondrial membrane potential percentage (NKG 2C + NKT.MMP low %) in monocytes (Mono.MMP low %）、NKG2A + lymphocyte) mitochondrial mass (NKG 2A + Lymph.MM) and average platelet volume (MPV).","assignee":"Competitive Peptide Biotechnology Hangzhou Co ltd; Hangzhjou Obstetrics & Gynecology Hospital; Ningbo Panpeptide Medical Laboratory Co ltd; Pan Peptide Biotechnology Zhejiang Co ltd","inventors":["郭鹏","董世雷","周青雪","李序帆"],"publication_date":"2026-01-02","filing_date":"2025-12-08","priority_date":"2025-12-08","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/30","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2135","G","G06","G06N","G06N20/00","G06N20/20","G","G16","G16B","G16B40/00","G","G16","G16H","G16H10/00","G16H10/20"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121260484A/en"},{"publication_number":"CN121259898A","title":"Facial asymmetry analysis method and system","abstract":"The application provides a face asymmetry analysis method and a face asymmetry analysis system, which relate to the technical field of intersection of computer vision and machine learning, in particular to a face asymmetry analysis method, comprising the steps of obtaining multi-frame face images to be evaluated; according to the multi-frame face image and a preset face key point detection algorithm, the face key points are detected, the eyebrow gradient, the eyebrow height and inner eye angle interval ratio, the mouth angle gradient and the mouth angle are calculated and obtained by utilizing the face key points to serve as static features, feature fusion is carried out on the dynamic features and the static features to obtain fusion features, and the fusion features are input into a second classification model which is trained in advance to obtain classification results. The application can improve the accuracy of the facial asymmetry analysis.","assignee":"Athena Eyes Co Ltd","inventors":["麻凯利","赵玮","谢晓彤","舒成成"],"publication_date":"2026-01-02","filing_date":"2025-12-08","priority_date":"2025-12-08","cpc_codes":["G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G06V40/168","G06V40/171","G","G06","G06N","G06N20/00","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G06V40/161","G06V40/165","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G06V40/172"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121259898A/en"},{"publication_number":"CN121255166A","title":"Policy DSL information generation method and related equipment","abstract":"The embodiment of the specification discloses a policy DSL information generation method and related equipment, and relates to the technical field of computers. The embodiment of the specification provides a hybrid intelligent cooperation architecture comprising an agent for rewriting, an agent for searching and an agent for generating, wherein the agent for rewriting analyzes a single initial instruction, expands the single initial instruction into a plurality of recall instructions, searches a strategy DSL information fragment matched with the recall instruction in a DSL knowledge base by the agent for searching, recalls a plurality of strategy DSL information fragments related to a strategy DSL information generating task, integrates the strategy DSL information fragments into a prompt word, generates grammar compliance based on the prompt word by the agent for generating strategy DSL information required by the initial instruction, and solves the problems of low efficiency and easiness in error of manually writing the strategy DSL information.","assignee":"Chongqing Ant Consumer Finance Co ltd","inventors":["黄天佑"],"publication_date":"2026-01-02","filing_date":"2025-12-05","priority_date":"2025-12-05","cpc_codes":["G","G06","G06F","G06F8/00","G06F8/30","G06F8/33","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/335","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121255166A/en"},{"publication_number":"CN121257500A","title":"Service data processing method and device based on online form","abstract":"The application discloses a business data processing method and device based on an online form, wherein the method comprises the steps of calling a preset online form module and initializing an online form file under the condition of receiving data processing requirements submitted by a user in a business system, integrating the preset online form module into the business system and providing an operation interface and a function identical to Excel, acquiring original business data of a plurality of subsystem modules in a background business database according to the data processing requirements and the preset standardized data interface, loading the original business data of each subsystem module into the online form file to obtain visual form data for display, and processing the visual form data in response to an operation instruction of the user on the displayed visual form data based on the operation interface and the function to generate target business data meeting the data processing requirements. Therefore, the application can improve the speed and efficiency of business data processing. And the overall quality and reliability of service data processing are improved.","assignee":"Puhuizhizao Technology Co ltd","inventors":["王克飞","徐超","应春红"],"publication_date":"2026-01-02","filing_date":"2025-12-05","priority_date":"2025-12-05","cpc_codes":["G","G06","G06F","G06F40/00","G06F40/10","G06F40/166","G06F40/177","G06F40/18","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481","G06F3/0482","G","G06","G06N","G06N20/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121257500A/en"},{"publication_number":"CN121259620A","title":"Building outer facade defect detection method and system based on bimodal image","abstract":"本申请属于智能检测与计算机视觉交叉技术领域，具体公开了一种基于双模态图像的建筑外立面缺陷检测方法及系统，该方法包括：基于建筑外立面的可见光图像和红外图像，进行对齐处理，获取对齐后的可见光图像和对齐后的红外图像；输入对齐后的可见光图像和对齐后的红外图像至双模态图像缺陷检测模型，获取双模态图像缺陷检测模型输出的建筑外立面的缺陷检测结果；其中，双模态图像缺陷检测模型是通过级联双流主干网络、跨模态注意力融合单元、语义掩码应用单元和检测头构建的。通过本申请，能够实现高效且准确地检测建筑外立面缺陷。 This application belongs to the interdisciplinary field of intelligent detection and computer vision, specifically disclosing a method and system for detecting defects in building facades based on bimodal images. The method includes: aligning visible light and infrared images of the building facade to obtain aligned visible light and infrared images; inputting the aligned visible light and infrared images into a bimodal image defect detection model to obtain the defect detection results of the building facade output by the bimodal image defect detection model; wherein the bimodal image defect detection model is constructed using a cascaded dual-stream backbone network, a cross-modal attention fusion unit, a semantic mask application unit, and a detection head. This application enables efficient and accurate detection of defects in building facades.","assignee":"Hubei Academy Of Architectural Sciences And Design Co ltd; HUBEI UNIVERSITY OF ECONOMICS","inventors":["刘文平","陈庆敏","李明磊","杨庆宜","姚欣蕊","何浩宇","戴君","易轶"],"publication_date":"2026-01-02","filing_date":"2025-12-05","priority_date":"2025-12-05","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/10","G06V20/176","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06T","G06T7/00","G06T7/30","G06T7/33","G06T7/337","G","G06","G06T","G06T7/00","G06T7/70","G06T7/73","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G06V10/765","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/70","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10004","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10048"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121259620A/en"},{"publication_number":"CN121256516A","title":"Emotion classification method based on audio-video self-adaptive perception fusion","abstract":"本发明公开了一种基于音视频自适应感知融合的情感分类方法，旨在解决多模态融合时，时序信息利用不充分、多模态数据分类过程中鲁棒性差的技术问题；构建一个包含语音分支、视频分支、融合分支的三分支网络结构；其中，语音分支、视频分支为单模态分支，在单模态分支中，通过时域特征挖掘层深入学习上下文依赖关系；在融合分支中，通过感知融合模块将单模态的时序信息引入到融合特征中；并利用模态有效性预测模块，动态评估各分支的可靠性并分配权重，获得最终的情感分类预测结果。本发明通过深化时序信息挖掘并自适应地权衡不同模态的有效性，有效缓解了因单一模态数据质量不佳而导致的模型性能下降问题，实现了更准确、更鲁棒的情感分类。 This invention discloses a sentiment classification method based on adaptive perceptual fusion of audio and video, aiming to solve the technical problems of insufficient utilization of temporal information and poor robustness in multimodal data classification during multimodal fusion. It constructs a three-branch network structure including a speech branch, a video branch, and a fusion branch. The speech and video branches are single-modal branches, in which contextual dependencies are deeply learned through a temporal feature mining layer. In the fusion branch, the temporal information of the single modality is introduced into the fusion features through a perceptual fusion module. A modality validity prediction module is used to dynamically evaluate the reliability of each branch and assign weights to obtain the final sentiment classification prediction result. This invention effectively alleviates the problem of model performance degradation caused by poor quality of single-modal data by deepening temporal information mining and adaptively balancing the validity of different modalities, achieving more accurate and robust sentiment classification.","assignee":"Nanjing University of Information Science and Technology","inventors":["吴泽远","孙玉宝"],"publication_date":"2026-01-02","filing_date":"2025-12-05","priority_date":"2025-12-05","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","Y","Y02","Y02D","Y02D10/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121256516A/en"},{"publication_number":"CN121236246A","title":"A digital mouth shape prediction system and method based on variational autoencoder in smart agriculture","abstract":"The application provides a digital human mouth type prediction system and a digital human mouth type prediction method based on a variation self-encoder in intelligent agriculture, which relate to the technical field of artificial intelligence, and the application extracts multidimensional voice characteristics and fuses and generates voice representation by receiving voice signals of intelligent agriculture scenes; the method comprises the steps of inputting a voice representation into a model, decoding the voice representation into a mouth-shaped key point sequence, combining agricultural proper nouns in the input voice with a preset mouth-shaped template, carrying out semantic enhancement on the mouth-shaped key point sequence, simultaneously inhibiting mouth-shaped jitter and abnormal frames, mapping the enhanced mouth-shaped key point sequence to a three-dimensional digital human face model, combining face parameters of a target user to convert the face parameters into facial soft tissue deformation data, calculating skeleton node positions and rotation parameters according to a preset corresponding relation between the mouth-shaped and the skeleton, and generating a digital human mouth-shaped animation synchronous with the input voice through combined rendering, so that stable and accurate voice synchronous digital human mouth shapes can be generated in an intelligent agricultural scene.","assignee":"Li'an Zhitong Beijing Technology Co ltd","inventors":["潘航","袁海杰"],"publication_date":"2025-12-30","filing_date":"2025-12-04","priority_date":"2025-12-04","cpc_codes":["G","G06","G06T","G06T13/00","G06T13/20","G06T13/40","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G10","G10L","G10L21/00","G10L21/06","G10L21/10","G","G10","G10L","G10L21/00","G10L21/06","G10L21/10","G10L2021/105","Y","Y02","Y02A","Y02A40/00","Y02A40/10"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121236246A/en"},{"publication_number":"CN121234783A","title":"A Machine Learning-Based Vacuum Induction Rapid Remelting Centrifugal Casting Method and Equipment","abstract":"本发明属于真空感应重熔离心铸造工艺技术领域，具体涉及基于机器学习的真空感应快速重熔离心铸造方法及设备，通过重熔仿真实验和实验数据获得包含不同重熔工艺参数和重熔特征参数的数据库，基于数据库训练得到预测真空感应快速重熔工艺参数的重熔机器学习模型，根据目标离心铸造工艺参数设计目标重熔特征参数，利用训练后的重熔机器学习模型，输出对应的最优重熔工艺参数并进行真空感应快速重熔离心铸造；本发明集成了仿真实验和机器学习实验的优点，降低生产成本、产品设计和试制的周期，实现了工艺参数的最优化，为安全生产保证了双层保障，进一步提高了生产效率。 This invention belongs to the field of vacuum induction remelting centrifugal casting technology, specifically involving a machine learning-based method and equipment for rapid vacuum induction remelting centrifugal casting. It obtains a database containing different remelting process parameters and remelting characteristic parameters through remelting simulation experiments and experimental data. Based on this database, a remelting machine learning model is trained to predict the rapid vacuum induction remelting process parameters. Target remelting characteristic parameters are designed according to the target centrifugal casting process parameters. Using the trained remelting machine learning model, the corresponding optimal remelting process parameters are output, and rapid vacuum induction remelting centrifugal casting is performed. This invention integrates the advantages of simulation experiments and machine learning experiments, reducing production costs, shortening product design and trial production cycles, optimizing process parameters, providing double-layer protection for safe production, and further improving production efficiency.","assignee":"Beihang University","inventors":["张花蕊","程颖","张虎","徐惠彬"],"publication_date":"2025-12-30","filing_date":"2025-12-02","priority_date":"2025-12-02","cpc_codes":["B","B22","B22D","B22D13/00","B22D13/12","F","F27","F27B","F27B14/00","F27B14/04","F","F27","F27B","F27B14/00","F27B14/08","F27B14/20","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N20/00","F","F27","F27B","F27B14/00","F27B14/04","F27B2014/045"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121234783A/en"},{"publication_number":"CN121236399A","title":"A method for segmenting complex shapes based on multidimensional information","abstract":"The invention relates to the technical field of image processing, in particular to a complex morphological target segmentation method based on multidimensional information guiding. And acquiring a target image to be segmented. Inputting the target image into a trained target segmentation model, and extracting a first feature map of the target image through a convolution layer of the target segmentation model. And inputting the first characteristic diagram into the encoder, and sequentially passing through a first coding layer, a second coding layer, a third coding layer and a fourth coding layer of the encoder to obtain a second characteristic diagram of the target image. And inputting the second feature map into a multidimensional information extraction module of the target segmentation model to obtain a third feature map of the target image. And inputting the third characteristic diagram, the output of the first coding layer, the output of the second coding layer, the output of the third coding layer and the first characteristic diagram into a decoder to obtain the output of the decoder, and inputting the output of the decoder into a segmentation layer to obtain the target segmentation image. The method is favorable for accurately positioning the position of the segmented target and keeping the definition and detail of the edge of the segmented target.","assignee":"Cas Intelligent Network Technology Co ltd; China Automobile Research Institute Jiangsu Automotive Engineering Research Institute Co ltd; China Automotive Engineering Research Institute Co Ltd","inventors":["伍泽","胡玮明","李朝斌","李斌","周金应","张强","唐宇","李瑞洁"],"publication_date":"2025-12-30","filing_date":"2025-12-02","priority_date":"2025-12-02","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/20","G06V10/26","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V2201/00","G06V2201/07"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121236399A/en"},{"publication_number":"CN121239537A","title":"Automatic modulation recognition method, apparatus and equipment based on lightweight neural networks","abstract":"本申请涉及一种基于轻量级神经网络的自动调制识别方法、装置及设备，属于无线通信技术领域。所述方法包括：构建了一种多流轻量级全局‑局部协同网络，在网络中首先采用多流输入架构，将并行的联合I/Q流、独立I流以及独立Q流作为输入融合处理，有效减少特征预处理冗余；其次引入反向残差模块，大幅精简参数量并提升特征提取效率；最后开发轻量级全局‑局部协同模块，将局部特征提取与全局关系建模深度融合后，利用分类模块进行调制类型识别分类。本方法能够更低的计算复杂度实现更高的调制类型识别准确率，且在低信噪比环境下展现出更强的鲁棒性。 This application relates to an automatic modulation identification method, apparatus, and device based on a lightweight neural network, belonging to the field of wireless communication technology. The method includes: constructing a multi-stream lightweight global-local cooperative network; firstly, employing a multi-stream input architecture to fuse parallel joint I/Q streams, independent I streams, and independent Q streams as inputs, effectively reducing feature preprocessing redundancy; secondly, introducing an inverse residual module to significantly simplify the number of parameters and improve feature extraction efficiency; and finally, developing a lightweight global-local cooperative module to deeply integrate local feature extraction with global relationship modeling, and then using a classification module for modulation type identification and classification. This method achieves higher modulation type identification accuracy with lower computational complexity and exhibits stronger robustness in low signal-to-noise ratio environments.","assignee":"National University of Defense Technology","inventors":["秦启儿","唐麒","周双敏","纪澎善","马东堂","魏急波"],"publication_date":"2025-12-30","filing_date":"2025-12-02","priority_date":"2025-12-02","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2431","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","H","H04","H04L","H04L27/00","H04L27/0012"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121239537A/en"},{"publication_number":"CN121234165A","title":"Acoustic emission signal classification methods, systems, devices, and media based on contrastive learning","abstract":"The invention discloses a method, a system, equipment and a medium for classifying acoustic emission signals based on contrast learning, which belong to the technical field of deep learning and signal processing, and the technical problem to be solved by the invention is how to improve the accuracy and generalization capability of acoustic emission signal classification, and the adopted technical scheme is as follows: constructing a plurality of groups of samples, namely collecting original acoustic emission signals with grade labels, preprocessing the collected original acoustic emission signals with grade labels, obtaining feature vectors, setting a sample number threshold value, and carrying out physical reasonable disturbance enhancement on a minority of samples with the sample number lower than the set threshold value; and regarding each sample as an anchor point, constructing 1+N multi-element group samples covering all categories for each anchor point, wherein N is the total number of categories obtained according to the grade labels, calculating the comparison loss of the hierarchical multi-element groups, namely extracting the characteristics of the multi-element group samples through a deep learning model to obtain characteristic embedded vectors, and training and classifying the deep learning model.","assignee":"Inspur Enterprise Cloud Technology Shandong Co ltd","inventors":["李志华","豆豪磊","王臻","刘傲","张积磊","肖港华","宋晨旭","朱翔宇","范成城"],"publication_date":"2025-12-30","filing_date":"2025-12-02","priority_date":"2025-12-02","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121234165A/en"},{"publication_number":"CN121237112A","title":"A method for human voice noise reduction in computer room based on 2D-TASNet","abstract":"本发明属于机房降噪技术领域，具体涉及一种基于2D‑TASNet的机房场内人声降噪方法，预处理：对混合信号进行短时傅里叶变换，将一维时域信号转换为二维时频谱；特征分解步骤：采用稀疏非负矩阵因式分解对时频谱的幅度谱进行特征提取；特征聚类：对SNMF分解得到的矩阵进行K‑Means聚类，使用纯净人声时频谱的质心作为聚类中心，以增强分离效果的鲁棒性；人声分离步骤：调整原生TASNet为2D‑TASNet，以K‑Means聚类得到的特征作为输入，训练2D‑TASNet得到目标掩码，并用此掩码重构分离的人声。噪声和人声的频率特征不同，2D‑TASNet的结构侧重于挖掘各种时频的潜在特征，这样的方式既能利用人声噪声各自的时频特性和非线性混合，又能保持神经网络的精度优势。 This invention belongs to the field of noise reduction technology in computer rooms, specifically involving a human voice noise reduction method based on 2D-TASNet. The method includes: preprocessing: performing a short-time Fourier transform on the mixed signal to convert the one-dimensional time-domain signal into a two-dimensional time-frequency spectrum; feature decomposition: extracting features from the amplitude spectrum of the time-frequency spectrum using sparse non-negative matrix factorization; feature clustering: performing K-Means clustering on the matrix obtained from SNMF decomposition, using the centroid of the pure human voice time-frequency spectrum as the cluster center to enhance the robustness of the separation effect; human voice separation: adjusting the original TASNet to a 2D-TASNet, using the features obtained from K-Means clustering as input to train the 2D-TASNet to obtain a target mask, and using this mask to reconstruct the separated human voice. Noise and human voice have different frequency characteristics. The structure of 2D-TASNet focuses on mining various potential time-frequency features. This approach can utilize the respective time-frequency characteristics and nonlinear mixing of human voice and noise while maintaining the accuracy advantage of neural networks.","assignee":"Sichuan Huakun Zhenyu Intelligent Technology Co ltd","inventors":["戢芳","徐洋"],"publication_date":"2025-12-30","filing_date":"2025-12-02","priority_date":"2025-12-02","cpc_codes":["G","G10","G10L","G10L21/00","G10L21/02","G10L21/0208","G10L21/0216","G10L21/0232","G","G06","G06F","G06F18/00","G06F18/20","G06F18/23","G06F18/232","G06F18/2321","G06F18/23213","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G10","G10L","G10L21/00","G10L21/02","G10L21/0208","G10L21/0216","G","G10","G10L","G10L21/00","G10L21/02","G10L21/0208","G10L21/0216","G10L21/0224","G","G10","G10L","G10L21/00","G10L21/02","G10L21/0208","G10L21/0264","G","G10","G10L","G10L25/00","G10L25/27","G","G10","G10L","G10L25/00","G10L25/27","G10L25/30"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121237112A/en"},{"publication_number":"CN121233637A","title":"A method and device for intelligent selection of industrial instruments through collaboration between RAG and expert systems.","abstract":"The invention provides an intelligent type selection method and device for an industrial instrument by cooperation of an RAG and an expert system, and belongs to the technical field of artificial intelligence and industrial automation intersection. The invention divides L instrument parameters with the selectable item number K of each parameter into a core parameter set with M less than or equal to 6 parameters and an attribute parameter set with N=L-M parameters, wherein the core parameter set defines a basic model, realizes the dynamic binding of the basic model stored in a table structure and the attribute parameter stored in a dictionary structure, and the RAG generates accurate parameter suggestion based on expert knowledge by a user according to the natural language query parameter problem, and the expert system verifies the parameter accuracy and cross-parameter compatibility in real time and provides correction suggestion when in conflict. The model selection process is two-stage matching, namely, bitmap index locking the basic model corresponding to the core parameter, and hash searching and verifying the attribute parameter. The instrument model selection method provided by the invention not only obtains the result of the instrument, but also is a solution for meeting the application requirements of the instrument.","assignee":"Zhuoran Tiangong Automation Instrument Beijing Co ltd","inventors":["周军"],"publication_date":"2025-12-30","filing_date":"2025-12-02","priority_date":"2025-12-02","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2457","G","G06","G06F","G06F16/00","G06F16/20","G06F16/22","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/242","G06F16/2428","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/242","G06F16/243","G","G06","G06F","G06F16/00","G06F16/30","G06F16/31","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121233637A/en"},{"publication_number":"CN121235419A","title":"A Method and System for Scheduling Construction Resources for Stadium Buildings Based on a Large Model","abstract":"本发明公开了基于大模型的体育场建筑建设施工资源调度方法及系统，通过构建时空特征矩阵；将时空特征矩阵输入至混合预测模型中，利用GCN融合图卷积网络学习工序之间的复杂拓扑依赖和级联影响，通过LSTM长短期记忆网络学习每个工序进度的时间序列模式并融合时空特征矩阵，输出的施工完成概率指数；基于所述施工完成概率指数驱动多智能体强化学习MARL框架，以实现全局目标为准则设计，通过模拟仿真预测资源需求曲线；根据所述资源需求曲线生成施工资源调度方案，至少包括资源的分配量、调度时间和移动路径。可直接指导现场作业，使得资源调度系统能够动态响应施工现场变化。 This invention discloses a method and system for scheduling construction resources for stadium buildings based on a large model. It constructs a spatiotemporal feature matrix; inputs this matrix into a hybrid prediction model; utilizes a graph convolutional network (GCN) to learn the complex topological dependencies and cascading effects between construction processes; and employs an LSTM (Long Short-Term Memory) network to learn the time-series patterns of each process's progress, fusing the spatiotemporal feature matrix to output a construction completion probability index. Based on this probability index, a multi-agent reinforcement learning (MARL) framework is driven, designed with the global objective as the criterion. Resource demand curves are predicted through simulation; and a construction resource scheduling scheme is generated based on the resource demand curves, including at least the resource allocation amount, scheduling time, and movement path. This directly guides on-site operations, enabling the resource scheduling system to dynamically respond to changes at the construction site.","assignee":"Middle East Infrastructure Technology Group Co ltd","inventors":["刘雨洲","李峰","刘亚峰","李超","贺叶发","郭炜芬"],"publication_date":"2025-12-30","filing_date":"2025-12-02","priority_date":"2025-12-02","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06312","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06315","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06316","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/067","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087","G","G06","G06Q","G06Q50/00","G06Q50/08","G","G06","G06F","G06F2123/00","G06F2123/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121235419A/en"},{"publication_number":"CN121235130A","title":"A Method and System for Accelerating Inference Based on Dynamic Sparsity of Large Language Models","abstract":"本发明公开了一种基于动态稀疏性的大语言模型推理加速方法及系统，本发明方法包括针对原始的目标大语言模型，在其支持结构动态精简的网络模块原有的主计算路径上增加并行的旁路预测路径，在旁路预测路径中嵌入选择激活的预测器，所述预测器用于根据该网络模块的输入向量来生成需激活的网络子模块，从而得到支持稠密模式和稀疏模式两种工作模式的目标大语言模型；在需要执行稀疏模式时，激活旁路预测路径中嵌入的预测器以获得快速推理结果；在需要执行稠密模式时，关闭旁路预测路径中嵌入的预测器以获得全面推理结果。本发明旨在解决大语言模型推理过程中的显存占用过高与时间消耗过大的问题，实现计算效率与资源消耗的优化平衡。 This invention discloses a method and system for accelerating inference in large language models based on dynamic sparsity. The method involves adding parallel side-path prediction paths to the original main computation path of the network modules supporting dynamic structural simplification in the original target large language model. Selective predictors are embedded in these side-path prediction paths. These predictors generate network sub-modules to be activated based on the input vector of the network module, thus obtaining a target large language model supporting both dense and sparse modes. When sparse mode is required, the predictors embedded in the side-path prediction paths are activated to obtain fast inference results; when dense mode is required, the predictors embedded in the side-path prediction paths are deactivated to obtain comprehensive inference results. This invention aims to solve the problems of excessive memory usage and time consumption during large language model inference, achieving an optimized balance between computational efficiency and resource consumption.","assignee":"National University of Defense Technology","inventors":["蹇松雷","杨莹","余杰","李宝","张建锋","丁滟","谭霜","王怡琦","郭勇","王晓川"],"publication_date":"2025-12-30","filing_date":"2025-12-02","priority_date":"2025-12-02","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121235130A/en"},{"publication_number":"KR20250178708A","title":"Automated Data Product Management System and Method Using an Extended DCAT-Based Integrated Data Model Data Control Algorithm","abstract":"본 발명은 통합 데이터모델 기반의DCAT(데이터 카탈로그 표준)을 확장하여 데이터상품의 생성, 품질평가, 버전관리, 가격정책 및 유통·정산을 자동화하는 데이터상품 관리 시스템 및 방법에 관한 것이다. 본 발명은 Δhash 기반 변경감지 및 머신러닝 기반 품질평가 알고리즘을 통합하여, 데이터상품의 신뢰성·무결성·상호운용성을 기술적으로 보장한다. 특히, Δhash 기반 자동 버전검증은 기존 데이터유통시스템에서 관리자가 수행하던 수작업 검증 절차를 대체하여, 데이터 무결성 검증의 효율성과 정확성을 향상시키며, 데이터 품질에 따른 가격정책과 정산을 실시간으로 자동 수행할 수 있는 효과가 있다. 이로써 데이터 유통 전 주기의 자동화 및 품질 중심의 지능형 데이터거래 체계를 구현할 수 있다. The present invention relates to a data product management system and method that automates the creation, quality assessment, version management, pricing policy, and distribution/settlement of data products by extending the Data Catalog Standard (DCAT) based on an integrated data model. The present invention technically guarantees the reliability, integrity, and interoperability of data products by integrating a Δhash-based change detection and a machine learning-based quality assessment algorithm. In particular, Δhash-based automatic version verification replaces the manual verification process performed by administrators in existing data distribution systems, thereby improving the efficiency and accuracy of data integrity verification and enabling automatic real-time execution of pricing policies and settlements based on data quality. This will enable the implementation of an intelligent data transaction system that focuses on automation and quality throughout the entire data distribution cycle.","assignee":"은진영","inventors":["은진영"],"publication_date":"2025-12-29","filing_date":"2025-11-22","priority_date":"2025-11-22","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/25","G","G06","G06F","G06F16/00","G06F16/20","G06F16/21","G06F16/211","G","G06","G06F","G06F16/00","G06F16/20","G06F16/21","G06F16/215","G","G06","G06F","G06F16/00","G06F16/20","G06F16/21","G06F16/219","G","G06","G06F","G06F16/00","G06F16/20","G06F16/22","G06F16/2228","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2452","G","G06","G06F","G06F16/00","G06F16/80","G06F16/84","G06F16/86","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F21/00","G06F21/10","G06F21/16","G","G06","G06F","G06F21/00","G06F21/60","G06F21/64","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0206","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0278","H","H04","H04L","H04L9/00","H04L9/50"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250178708A/en"},{"publication_number":"KR102905017B1","title":"Ai-based system for detecting dangerous situation an abnormal behaviors in fitness centers and providing real-time remote intervention","abstract":"피트니스 센터 내 위험을 감지 및 원격 제어 서비스를 제공하기 위한 서비스 제공 장치는, 네트워크 인터페이스, 상기 네트워크 인터페이스와 작동적으로(operatively) 연결된 적어도 하나의 프로세서, 및 상기 적어도 하나의 프로세서와 작동적으로 연결된 적어도 하나의 메모리를 포함하고, 상기 적어도 하나의 메모리는, 실행시, 상기 적어도 하나의 프로세서가, 상기 대상 공간에 설치된 카메라로부터 제1 서비스 데이터를 획득하고, 상기 대상 공간에 설치된 바이오 센서로부터 제2 서비스 데이터를 획득하고, 상기 제1 서비스 데이터 및 상기 제2 서비스 데이터를, 컴퓨터 비전(computer vision) 및 자연어 처리(national language process)에 기반한 대상 모델에 적용하여, 상기 대상 공간에 포함된 사용자 및 외부 장치의 이상 상황 여부를 식별하는 결과 데이터를 획득하고, 상기 결과 데이터를 통해 상기 대상 공간에 이상 상황이 발생한 것으로 식별한 것에 기반하여, 관제 서버에 알림을 제공하도록 야기하는 명령어들을 저장할 수 있다. A service providing device for detecting risks in a fitness center and providing a remote control service includes a network interface, at least one processor operatively connected to the network interface, and at least one memory operatively connected to the at least one processor, wherein the at least one memory can store commands that, when executed, cause the at least one processor to obtain first service data from a camera installed in the target space, obtain second service data from a biosensor installed in the target space, apply the first service data and the second service data to a target model based on computer vision and natural language processing to obtain result data identifying whether a user and an external device included in the target space are in an abnormal state, and provide a notification to a control server based on the identification that an abnormal state has occurred in the target space through the result data.","assignee":"주식회사 서플라이스","inventors":["천인섭","김준동"],"publication_date":"2025-12-29","filing_date":"2025-07-09","priority_date":"2025-07-09","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06F","G06F40/00","G06F40/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/50","G06V20/52","G","G06","G06V","G06V40/00","G06V40/20","G06V40/23","G","G08","G08B","G08B13/00","G08B13/18","G08B13/189","G08B13/194","G08B13/196","G","G08","G08B","G08B21/00","G08B21/18","G08B21/182"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102905017B1/en"},{"publication_number":"KR102903147B1","title":"Artificail intelligence-based underground pipeline monitoring system using pipeline trajectory device and operating method therefor","abstract":"본 일 실시 예에 따른 관로투입장치를 이용한 인공지능 기반 지중관로 모니터링 시스템을 위한 동작 방법에 있어서, 관로투입장치(30)가 지중관로 내부를 주행하면서 카메라 모듈로부터 획득한 관내 영상 데이터를 수신하는 단계; 관내 영상 데이터를 기반으로 적어도 하나의 관내 크랙을 탐지하는 단계; 적어도 하나의 크랙이 탐지될 때, 적어도 하나의 관내 크랙과 상응하는 적어도 하나의 크랙탐지구역을 위한 크랙좌표정보 및 크랙유형정보를 생성하는 단계; 관내 영상 데이터를 기반으로 적어도 하나의 크랙탐지구역과 연관된 복수의 이미지 프레임을 추출하는 단계; 및 추출된 복수의 이미지 프레임을 기반으로 하이라이트 영상정보를 생성하는 단계를 포함한다. In an operating method for an artificial intelligence-based underground pipeline monitoring system using a pipeline insertion device according to an embodiment of the present invention, the method comprises: receiving internal image data acquired from a camera module while the pipeline insertion device (30) travels inside an underground pipeline; detecting at least one internal crack based on the internal image data; generating crack coordinate information and crack type information for at least one crack detection zone corresponding to at least one internal crack when at least one crack is detected; extracting a plurality of image frames associated with at least one crack detection zone based on the internal image data; and generating highlight image information based on the extracted plurality of image frames.","assignee":"(주) 대륙전설","inventors":["홍종경"],"publication_date":"2025-12-29","filing_date":"2025-07-01","priority_date":"2025-07-01","cpc_codes":["G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/95","G01N21/954","F","F16","F16L","F16L55/00","F16L55/26","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G06T7/001","F","F16","F16L","F16L2101/00","F16L2101/30","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G01N2021/8854","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G01N2021/8887"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102903147B1/en"},{"publication_number":"KR102905296B1","title":"Automated processing method, device and system for digital integrated civil complaint processing using generative artificial intelligence model","abstract":"일실시예에 따른 장치는, 민원인의 단말로부터 민원 데이터를 수신하고, 수신된 민원 데이터를 분석하여 민원 카테고리 및 민원 주요 키워드를 포함하는 민원 요약 정보를 추출하고, 과거 민원 처리 사례, 민원 처리 매뉴얼 및 민원 응답 기본 형식이 저장된 민원 데이터베이스를 기반으로 학습된 생성형 인공지능 모델을 통해, 민원 데이터에 대한 민원 응답 문안을 생성하고, 민원 데이터의 처리를 위한 관계 기관의 단말에 민원 요약 정보 및 민원 응답 문안을 전송할 수 있다. According to one embodiment, a device receives complaint data from a complaint terminal, analyzes the received complaint data to extract complaint summary information including complaint categories and complaint main keywords, generates a complaint response text for the complaint data through a generative artificial intelligence model learned based on a complaint database storing past complaint handling cases, complaint handling manuals, and basic complaint response formats, and transmits the complaint summary information and complaint response text to a terminal of a relevant agency for processing the complaint data.","assignee":"영원아이앤에스(주)","inventors":["김상환"],"publication_date":"2025-12-29","filing_date":"2025-06-30","priority_date":"2025-06-30","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102905296B1/en"},{"publication_number":"LU602278B1","title":"Method for computation offloading and resources allocation based on deep reinforcement learning","abstract":"The invention discloses a method for computation offloading and resources allocation based on deep reinforcement learning, namely a Double Deep Q-Learning, to deal with the problem of global cost minimization. According to the method, the optimal strategy for computation offloading and resources allocation can be obtained under the time-varying channel state and random task arrival environment, and a deep neural network is used as an optimizer of a value function, which can reduce the dimension disaster caused by a high-dimensional state space and improve the convergence speed.","assignee":"Changchun Inst Tech","inventors":["Hongchang Ke"],"publication_date":"2025-12-29","filing_date":"2025-06-26","priority_date":"2025-06-26","cpc_codes":["G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5061","G06F9/5072","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/48","G06F9/4806","G06F9/4843","G06F9/4881","G06F9/4887","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5061","G06F9/5066","G","G06","G06N","G06N3/00","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/5017","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/509","H","H04","H04L","H04L67/00","H04L67/50","H04L67/56","H04L67/59"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU602278B1/en"},{"publication_number":"LU602264B1","title":"Ai-based cross-scale multi-omics system for immunotherapy efficacy prediction and prognostic assessment in gastric cancer and colorectal cancer","abstract":"The present invention relates to the technical field of medical artificial intelligence, and provides an artificial intelligence (AI)-based cross-scale multi-omics system for immunotherapy efficacy prediction and prognostic assessment in gastric cancer and colorectal cancer, including a multi-modal data acquisition module, a data fusion and feature extraction module, a cross-scale integrated learning module, a dynamic prediction update module, an interpretability analysis module, and a treatment recommendation module. The AI-based cross-scale multi-omics system for immunotherapy efficacy prediction and prognostic assessment in gastric cancer and colorectal cancer of the present invention breaks through technical bottlenecks of existing single-omics models, including insufficient depth of fusion, poor timeliness of static prediction, and weak interpretability of black-box models. Through cross-scale data fusion and dynamic weight optimization mechanisms, it realizes end-to-end intelligence from multi-omics feature extraction to clinical decision-making, providing precise and dynamic efficacy prediction and prognostic assessment tools for immunotherapy of gastric cancer and colorectal cancer.","assignee":"The First Peoples Hospital Of Zhenjiang City; Jiangsu Province Suqian Hospital","inventors":["Changfeng Man","Xiaoyan Wang","Yu Fan","Dandan Gong"],"publication_date":"2025-12-29","filing_date":"2025-06-26","priority_date":"2025-06-26","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/20","G","G06","G06N","G06N3/00","G","G16","G16B","G16B40/00","G16B40/20","G","G16","G16H","G16H20/00","G16H20/10","G","G16","G16H","G16H40/00","G16H40/60","G16H40/63","G","G16","G16H","G16H50/00","G16H50/70"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU602264B1/en"},{"publication_number":"LU602240B1","title":"Multi-omics data analysis system for predicting the efficacy of immunotherapy in head and neck tumors","abstract":"A multi-omics data analysis system for predicting the efficacy of immunotherapy in head and neck tumors, wherein it comprises: a data acquisition module, configured to obtain patient genomic, transcriptomic, proteomic, and clinical data. Data integration for efficiency improvement: multi-omics data such as genomics and transcriptomics are collected and deeply integrated to mine multi-level biological information; compared with traditional single-data prediction, this significantly improves the accuracy and reliability of immunotherapy efficacy prediction. Algorithmic performance optimization: key features are selected using SHAP values and leave-one-out methods; with the aid of attention mechanisms and ensemble models, the nonlinear fitting and generalization capabilities of the model are enhanced, the risk of overfitting is reduced, and the prediction performance is optimized.","assignee":"The First Affiliated Hospital Of Univ Of South China","inventors":["Guicheng He"],"publication_date":"2025-12-29","filing_date":"2025-06-25","priority_date":"2025-06-25","cpc_codes":["G","G16","G16H","G16H20/00","G16H20/10","G","G06","G06N","G06N3/00","G","G16","G16B","G16B40/00","G16B40/20","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/50","G","G16","G16H","G16H50/00","G16H50/70"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU602240B1/en"},{"publication_number":"LU602238B1","title":"Optimization identification method for svg parameter based on sac deep reinforcement learning","abstract":"The invention discloses an optimization identification method for SVG parameter based on SAC deep reinforcement learning, which comprises the following steps: Establish an equivalent mathematical model of SVG accessing infinite system; calculate and screen the reactive power trajectory sensitivity of each parameter by using the perturbation method; establish the environment of SAC based on BPA; build a SAC intelligent agent; start SVG parameter identification training to get the result. According to the present invention, the parameters of the SVG controller are identified by using the SAC model, so that the time consumption is less, the accuracy of the parameter prediction result is high, and the identification efficiency is high.","assignee":"Hangzhou Dianzi Univ Information Engineering College","inventors":["Xingchi Fan","Zeyu Zhong","Shihua Jin","Huimin Gao","Miao Guo","Jiayue Zhang"],"publication_date":"2025-12-29","filing_date":"2025-06-25","priority_date":"2025-06-25","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/30","G06F30/36","G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N3/00","H","H02","H02J","H02J1/00","G","G06","G06F","G06F2111/00","G06F2111/10"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU602238B1/en"},{"publication_number":"LU602243B1","title":"Method for identifying potential industrial sources of groundwater pollution","abstract":"Disclosed is a method for identifying an industry classification of a potential industrial source and characteristic pollutants of the potential industrial source, wherein the method for identifying an industry classification of a potential industrial source comprises: acquiring information point data of a target potential industrial source; determining feature words of the information point data and feature values of the feature words according to a preset semantic lexicon, preset industry summary information and the information point data; and determining the industry classification to which the target potential industrial source belongs according to a preset industry classification prediction model and the feature values. Through implementing the present invention, the obtained feature values can effectively avoid interference of meaningless words, such that the industry classification to which the target potential industrial source belongs obtained from identification is more accurate.","assignee":"Chinese Academy Of Env Planning","inventors":["Guoxin Huang"],"publication_date":"2025-12-29","filing_date":"2025-06-25","priority_date":"2025-06-25","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/06","G","G06","G06N","G06N3/00","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU602243B1/en"},{"publication_number":"KR20250179106A","title":"Artificial intelligence counseling method for sequentially performing ethics-based data filtering and user-tailored data provision","abstract":"본 발명의 실시 예에 따른 인공지능 모델 기반의 상담 서버의 동작 방법은 사용자에 대한 설문 정보를 기반으로 해당 사용자의 유형을 식별하는 단계, 인공지능 모델 기반의 복수의 상담 모델 중 식별된 사용자 유형에 대응되는 상담 모델을 선택하는 단계, 상기 사용자 유형에 기초하여, 텍스트 데이터에 대한 필터링 기준을 설정하는 단계, 상기 상담 모델이 상담 응답을 생성하기 위해 참조하는 전방 데이터를 상기 필터링 기준에 따라 필터링하는 단계 및 필터링된 데이터와 사용자 특성에 대응하여 생성된 보완 데이터를 기반으로 상기 상담 모델이 상담 응답을 생성하는 단계를 포함하여 구성될 수 있다. An operation method of an artificial intelligence model-based consultation server according to an embodiment of the present invention may be configured to include a step of identifying a type of a user based on questionnaire information about the user, a step of selecting a consultation model corresponding to the identified user type from among a plurality of consultation models based on the artificial intelligence model, a step of setting a filtering criterion for text data based on the user type, a step of filtering forward data referenced by the consultation model to generate a consultation response according to the filtering criterion, and a step of generating a consultation response by the consultation model based on the filtered data and supplementary data generated corresponding to user characteristics.","assignee":"문만기","inventors":["문만기"],"publication_date":"2025-12-29","filing_date":"2025-06-20","priority_date":"2024-06-20","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G16","G16H","G16H10/00","G16H10/20","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/70"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250179106A/en"},{"publication_number":"CN121215148A","title":"A Disease Prediction Method and System Based on Medical Invoices and False Labeling Mechanism","abstract":"本发明公开了一种基于医疗票据与伪标签机制的病种预测方法及系统，获取包含药品清单、就诊科室及患者画像信息的医疗票据集合后，经标准化预处理后生成标准化票据数据；将标准化票据数据分别输入动态推理知识库和预训练的大语言模型，通过规则推理引擎生成第一伪标签病种集合，同时通过语义理解提示词模板引导生成第二伪标签病种集合；融合两个伪标签病种集合得到融合伪标签，以此为目标训练多标签病种分类模型，最终实现对待预测医疗票据数据的病种预测。本发明通过融合规则推理与大语言模型的双重伪标签生成机制，解决了医疗票据数据缺乏真实病种标签的问题，提升了病种预测的准确性和可靠性，为医疗票据数据的深度利用提供了可靠的解决方案。 This invention discloses a method and system for disease prediction based on medical invoices and a pseudo-labeling mechanism. After acquiring a set of medical invoices containing drug lists, visiting departments, and patient profiles, standardized invoice data is generated through standardized preprocessing. This standardized invoice data is then input into a dynamic reasoning knowledge base and a pre-trained large language model. A first set of pseudo-labeled diseases is generated through a rule-based reasoning engine, while a second set of pseudo-labeled diseases is generated guided by semantic understanding prompts. The two sets of pseudo-labeled diseases are then fused to obtain a fused pseudo-label. This fused pseudo-label is used to train a multi-labeled disease classification model, ultimately achieving disease prediction for the medical invoice data to be predicted. This invention solves the problem of the lack of real disease labels in medical invoice data by integrating rule-based reasoning and a large language model for dual pseudo-label generation, improving the accuracy and reliability of disease prediction and providing a reliable solution for the in-depth utilization of medical invoice data.","assignee":"Fujian Boss Software Co ltd","inventors":["王伙明","陈庸凯","王航宇","武宜婧","刘盈","郑淑凡"],"publication_date":"2025-12-26","filing_date":"2025-12-01","priority_date":"2025-12-01","cpc_codes":["G","G16","G16H","G16H10/00","G16H10/60","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H70/00","G16H70/20","G","G16","G16H","G16H70/00","G16H70/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121215148A/en"},{"publication_number":"CN121215279A","title":"Methods, systems, equipment, media, and products for assessing the risk level of biliary tract cancer.","abstract":"本申请公开了一种胆道癌疾病风险等级评估方法、系统、设备、介质及产品，涉及流行病学领域，该方法包括：获取不同人群的暴露因素以及结局变量；所述暴露因素为胆石症相关SNP；所述结局变量为胆道癌SNP；基于孟德尔随机化分析方法，对所述胆石症相关SNP进行筛选，确定工具变量；所述工具变量为筛选后的胆石症相关SNP；基于所述工具变量、所述暴露因素以及所述结局变量，评估胆石症与胆道癌之间的因果关系；所述因果关系包括强因果关系、中等因果关系以及弱因果关系；根据所述因果关系评估胆道癌疾病风险等级；所述胆道癌疾病风险等级包括高危等级、中危等级以及低危等级，本申请能够准确评估胆道癌疾病风险等级。 This application discloses a method, system, device, medium, and product for assessing the risk level of biliary tract cancer, relating to the field of epidemiology. The method includes: obtaining exposure factors and outcome variables from different populations; the exposure factors are gallstone-related SNPs; the outcome variables are biliary tract cancer SNPs; screening the gallstone-related SNPs based on Mendelian randomization analysis to determine instrumental variables; the instrumental variables are the screened gallstone-related SNPs; assessing the causal relationship between gallstones and biliary tract cancer based on the instrumental variables, the exposure factors, and the outcome variables; the causal relationship includes strong causation, moderate causation, and weak causation; and assessing the risk level of biliary tract cancer based on the causal relationship; the risk level of biliary tract cancer includes high-risk, intermediate-risk, and low-risk levels. This application can accurately assess the risk level of biliary tract cancer.","assignee":"DALIAN FRIENDSHIP HOSPITAL","inventors":["吕文才"],"publication_date":"2025-12-26","filing_date":"2025-12-01","priority_date":"2025-12-01","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/30","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G16","G16B","G16B20/00","G16B20/20","G","G16","G16H","G16H50/00","G16H50/70"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121215279A/en"},{"publication_number":"CN121213130A","title":"A Method and System for Vegetable Price Forecasting Based on Online Public Opinion","abstract":"The invention provides a vegetable price prediction method and a system based on network public opinion, and relates to the field of artificial intelligence, wherein the method comprises the steps of acquiring vegetable price network public opinion data of a preset historical period, and acquiring a price sequence of a plurality of candidate vegetable categories in the preset historical period; and inputting the price sequence of the predicted vegetable class and the vegetable price network public opinion data into a price prediction model to obtain the predicted price of the predicted vegetable class in a future period, wherein the predicted price is output by the price prediction model. The price prediction model comprises a sequential relation extraction module, a self-adaptive characteristic enhancement module and a full-connection layer which are sequentially connected. The vegetable planting method and the vegetable planting device have the advantages that the vegetable predicted price with higher accuracy is obtained, and the establishment of a vegetable planting plan is better guided.","assignee":"Huazhong Agricultural University","inventors":["李优柱","许爽","夏静波","李潇"],"publication_date":"2025-12-26","filing_date":"2025-12-01","priority_date":"2025-12-01","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0206","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121213130A/en"},{"publication_number":"CN121210934A","title":"AI-Assisted Green Lighting Scheme Evaluation and Intelligent Decision-Making Method and System","abstract":"The invention provides an AI-assisted green lighting scheme assessment and intelligent decision-making method and system, which relate to the field of intelligent lighting and comprise the steps of identifying lighting requirements of different space positions, dividing a region into a plurality of subareas and determining assessment weights; the method comprises the steps of constructing an index coupling relation map to determine an optimized sequence, identifying a common regulatable factor to generate a preliminary adjustment scheme, predicting variation through propagation algorithm and optimizing the scheme, and finally generating a regulation command for each subarea. The invention realizes the accurate distribution of illumination resources, improves the energy utilization efficiency, and meets the specific illumination requirements of different areas.","assignee":"Liangye Technology Group Co ltd","inventors":["王秀燕"],"publication_date":"2025-12-26","filing_date":"2025-11-28","priority_date":"2025-11-28","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/29","G","G05","G05B","G05B13/00","G05B13/02","G05B13/04","G05B13/048","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","H","H05","H05B","H05B47/00","H05B47/10","H05B47/155","H","H05","H05B","H05B47/00","H05B47/10","H05B47/165","H","H05","H05B","H05B47/00","H05B47/10","H05B47/175","H05B47/198","H05B47/1985"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121210934A/en"},{"publication_number":"CN121214239A","title":"Image Analysis-Based Defect Analysis Method, System, Equipment, and Medium for Transmission Line Tension Clamps","abstract":"The invention relates to the technical field of defect detection, and discloses a method, a system, equipment and a medium for analyzing the defects of a strain clamp of a power transmission line based on image analysis, wherein the method comprises the steps of acquiring multidimensional image data by adopting multi-view and multi-spectrum imaging equipment; the method comprises the steps of constructing a feature extraction framework to extract features, constructing a reinforcement learning model to realize analysis decision, constructing a target detection screening model comprising a plurality of pre-measuring heads, adopting an anchor frame generation strategy and a non-maximum suppression algorithm to detect and position the defect target of the analyzed image features, constructing a classification evaluation framework based on a strain clamp defect type feature database and a classification decision tree to judge the type and evaluate the severity of the detected defect target, and outputting and storing the defect analysis result in a preset data format and a preset storage strategy. The method solves the problems of insufficient extraction of complex defect characteristics and stiff decision mechanism of the traditional method, and comprehensively improves the detection capability and the operation and maintenance efficiency of the strain clamp defects.","assignee":"Guizhou Power Grid Co Ltd","inventors":["曾宪武","尤明洋","刘力皓","秦袁","王先锋","田地","李生福","简蓓","白洁","官涵宇","曾宪龙","唐猛","沙梦华","严莉","曾稳植","胡晓晨","何璇","吴雪冬","黄家瑞","蔡馨雅"],"publication_date":"2025-12-26","filing_date":"2025-11-28","priority_date":"2025-11-28","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/10","G06V20/194","G","G06","G06F","G06F16/00","G06F16/50","G06F16/58","G06F16/583","G","G06","G06F","G06F16/00","G06F16/50","G06F16/58","G06F16/583","G06F16/5854","G","G06","G06F","G06F16/00","G06F16/50","G06F16/58","G06F16/583","G06F16/5862","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06V","G06V10/00","G06V10/20","G06V10/25","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/40","G06V10/46","G06V10/469","G","G06","G06V","G06V10/00","G06V10/40","G06V10/54","G","G06","G06V","G06V10/00","G06V10/40","G06V10/58","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/761","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06V","G06V2201/00","G06V2201/07","Y","Y04","Y04S","Y04S10/00","Y04S10/50"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121214239A/en"},{"publication_number":"CN121211017A","title":"Methods, apparatus and systems for generating ground-penetrating radar training data","abstract":"The application relates to the technical field of geophysical exploration, and discloses a method for generating geological radar training data. The method comprises the steps of obtaining descriptors of physical form angles of underground structures corresponding to radar data in a first sample set, a second sample set and a third sample set, forming descriptors and sample labels corresponding to each sample into descriptor vectors, finally obtaining a plurality of descriptor vectors, dividing the descriptor vectors into one or more vector clusters according to similarity, fusing the descriptor vectors in each vector cluster to obtain fused descriptor vectors, mapping the fused descriptor vectors back to the underground structures according to fused descriptors in the fused descriptor vectors, obtaining radar data by the forward-modeling underground structures, and combining the fused sample labels in the radar data and the fused descriptor vectors to obtain training samples. By adopting the method, the objective degree of the finally obtained training sample can be improved. The application also discloses a device and a system for generating the geological radar training data.","assignee":"Tianjin Survey And Design Institute Group Co ltd","inventors":["杨金瑞","吴宇豪","王英杰","杨建博","陈奕达"],"publication_date":"2025-12-26","filing_date":"2025-11-28","priority_date":"2025-11-28","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121211017A/en"},{"publication_number":"CN121214502A","title":"A Multimodal Temporal Fusion Speech-Driven Gesture Generation Method","abstract":"本发明公开了一种多模态时序融合语音驱动手势生成方法，包括如下步骤：首先通过矢量量化变分自编码器模型学习手势运动的紧凑离散表示，为后续生成任务构建一个量化的潜在空间；然后对语音音频通过音频编码器提取音频特征；对说话人身份进行低维嵌入得到身份特征；将音频特征与历史手势序列在时序对齐后沿特征维拼接形成多模态初始表征；通过集成自注意力机制、以身份特征为条件的交叉注意力机制和Mamba模块的多模态时序融合模块进行深度特征融合；最终通过预训练的解码器重建目标手势序列。本发明解决多模态融合不充分、计算效率低和生成动作单一化的技术问题，能够生成自然流畅、个性化且满足实时交互要求的手势动画。 This invention discloses a multimodal temporal fusion speech-driven gesture generation method, comprising the following steps: First, a compact discrete representation of gesture motion is learned through a vector quantization variational autoencoder model, constructing a quantized latent space for subsequent generation tasks; then, audio features are extracted from the speech audio using an audio encoder; speaker identity is obtained through low-dimensional embedding; the audio features are concatenated with historical gesture sequences along the feature dimension after temporal alignment to form a multimodal initial representation; deep feature fusion is performed through a multimodal temporal fusion module integrating a self-attention mechanism, a cross-attention mechanism conditioned on identity features, and a Mamba module; finally, the target gesture sequence is reconstructed through a pre-trained decoder. This invention solves the technical problems of insufficient multimodal fusion, low computational efficiency, and monotonous generated actions, and can generate natural, smooth, personalized gesture animations that meet real-time interaction requirements.","assignee":"Jiangxi Normal University","inventors":["刘长红","万浩聪","赖宝泉","章威"],"publication_date":"2025-12-26","filing_date":"2025-11-28","priority_date":"2025-11-28","cpc_codes":["G","G10","G10L","G10L17/00","G10L17/02","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/254","G06F18/256","G","G06","G06F","G06F3/00","G06F3/01","G06F3/017","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/809","G06V10/811","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V40/00","G06V40/10","G06V40/107","G06V40/113","G","G06","G06V","G06V40/00","G06V40/20","G06V40/28","G","G10","G10L","G10L17/00","G10L17/06","G10L17/10","G","G10","G10L","G10L17/00","G10L17/22"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121214502A/en"},{"publication_number":"CN121211991A","title":"Performance analysis methods, apparatus, equipment and media for cemented riprap protective structures","abstract":"本发明提出了一种胶结抛石防护结构的性能分析方法、装置、设备及介质，涉及胶结抛石防护结构性能测试技术领域。针对现有技术对网格化胶结抛石研究存在宏观冲刷形态与材料性能演变脱节、短期试验结果无法反映数十年服役周期性能衰减的缺陷，导致无法建立材料退化与结构失效的定量关系，难以实现长期防护性能预测，本发明通过水槽模型试验获取冲刷坑体积演变数据，结合材料侵蚀试验建立强度衰减模型，将波流载荷作用下的冲刷损伤等效为材料侵蚀时间，基于等效损伤龄期实时修正材料强度，最终构建冲刷坑发展速率模型预测长期防护性能。 This invention proposes a performance analysis method, apparatus, equipment, and medium for cemented riprap protective structures, relating to the field of performance testing technology for cemented riprap protective structures. Addressing the shortcomings of existing technologies in studying gridded cemented riprap, such as the disconnect between macroscopic scour morphology and material performance evolution, and the inability of short-term test results to reflect performance degradation over decades of service, which prevents the establishment of a quantitative relationship between material degradation and structural failure and hinders long-term protective performance prediction, this invention obtains scour pit volume evolution data through flue model tests, establishes a strength attenuation model by combining it with material erosion tests, equates scour damage under wave-current loading to material erosion time, and corrects material strength in real time based on the equivalent damage age, ultimately constructing a scour pit development rate model to predict long-term protective performance.","assignee":"Tianjin Research Institute for Water Transport Engineering MOT","inventors":["王洋","陈汉宝","陈松贵","张启博","戈龙仔","段自豪","王依娜","朱颖涛"],"publication_date":"2025-12-26","filing_date":"2025-11-28","priority_date":"2025-11-28","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N3/00","G06N3/12","G06N3/126","G","G06","G06F","G06F2119/00","G06F2119/02","G","G06","G06F","G06F2119/00","G06F2119/14"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121211991A/en"},{"publication_number":"CN121213567A","title":"A Deep Learning-Based Intelligent Detection System for Surface Defects in Titanium Ingots","abstract":"本发明公开了一种基于深度学习的钛锭表面缺陷智能检测系统，包括图像采集与预处理模块，用于获取输入数据集；物理先验特征构建模块，用于输出物理先验引导图；改进型ConvNeXt V2主干特征提取模块，用于构建通过跨阶段特征映射生成最终输出特征；局部注意力特征增强模块，用于输出局部加权输出特征；先验引导门控融合模块，用于卷积注意力模块P‑CBAM产生加权融合输出特征；多尺度特征融合与检测模块，用于生成钛锭表面缺陷的预测结果；不确定度评估模块，用于输出的缺陷类别。本发明实现了对钛锭表面多类型缺陷的高精度智能识别，适用于钛材加工生产线中缺陷筛查场景。 This invention discloses an intelligent detection system for titanium ingot surface defects based on deep learning, comprising: an image acquisition and preprocessing module for acquiring the input dataset; a physical prior feature construction module for outputting a physical prior guidance map; an improved ConvNeXt V2 backbone feature extraction module for constructing the final output features generated through cross-stage feature mapping; a local attention feature enhancement module for outputting locally weighted output features; a prior-guided gated fusion module for generating weighted fused output features using the convolutional attention module P-CBAM; a multi-scale feature fusion and detection module for generating prediction results of titanium ingot surface defects; and an uncertainty assessment module for outputting the defect category. This invention achieves high-precision intelligent identification of multiple types of defects on the surface of titanium ingots and is suitable for defect screening scenarios in titanium material processing production lines.","assignee":"Baoji Dali Wei Titanium Industry Co ltd","inventors":["杨帆"],"publication_date":"2025-12-26","filing_date":"2025-11-28","priority_date":"2025-11-28","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G06T7/0008","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","Y","Y02","Y02P","Y02P90/00","Y02P90/30"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121213567A/en"},{"publication_number":"CN121212031A","title":"A method for predicting reservoir production dynamics by fusing discrete gradient information","abstract":"The invention discloses a dynamic prediction method for oil deposit production integrating discrete gradient information, which belongs to the technical field of oil deposit development and artificial intelligence intersection and comprises the steps of building a heterogeneous oil deposit oil-water two-phase flow numerical simulation data set based on a numerical simulation method; the method comprises the steps of designing a dual-branch network structure to extract space and physical characteristics of input field data in parallel, designing a main network to conduct deep nonlinear modeling, constructing a high-efficiency pressure and saturation field prediction neural network model based on the dual-branch network structure and the main network, using space region observation points of partial time steps to participate in data item loss calculation in a model training stage, simultaneously introducing physical control equation residuals to serve as physical loss items in all time steps and whole space, and obtaining a trained high-efficiency pressure and saturation field prediction neural network model to realize high-precision prediction of a full-time pressure field and a saturation field.","assignee":"China University of Petroleum East China","inventors":["袁彬","熊浩男","张伟","吴一宁"],"publication_date":"2025-12-26","filing_date":"2025-11-28","priority_date":"2025-11-28","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/20","G06F30/28","G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06F","G06F2111/00","G06F2111/10","G","G06","G06F","G06F2113/00","G06F2113/08","G","G06","G06F","G06F2119/00","G06F2119/12","G","G06","G06F","G06F2119/00","G06F2119/14","Y","Y02","Y02A","Y02A10/00","Y02A10/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121212031A/en"},{"publication_number":"KR20250177392A","title":"Electronic device for performing depth-wise convolution, processing element for the same, and method for performing depth-wise convolution","abstract":"본 개시의 실시 예에 따른 전자 장치는 프로세서, 및 행 및 열 방향으로 배열되는 복수의 프로세싱 엘리먼트(processing element; PE)를 포함하는 시스톨릭 어레이(systolic array)를 포함하고, 상기 프로세서는, 가중치 매트릭스의 각 가중치 성분을 상기 시스톨릭 어레이의 대응하는 행에 맵핑시키고, 입력 매트릭스의 입력 열 벡터를 상기 시스톨릭 어레이의 행 방향으로 순차적으로 입력시키고, 기 설정된 수만큼의 입력 열 벡터가 상기 시스톨릭 어레이에 입력된 이후, 상기 시스톨릭 어레이에 다음 입력 열 벡터가 입력될 때, 상기 시스톨릭 어레이의 M행의 각 열에 입력된 입력 열 벡터의 입력 성분을 M+K행의 상기 각 열의 다음 열로 전달하고, 그리고 상기 시스톨릭 어레이의 모든 행에 입력 열 벡터가 입력된 것에 응답하여, 상기 가중치 매트릭스와 상기 입력 매트릭스의 연산을 통해 산출된 누적 합을 출력하되, 상기 M, K는 기 설정된 자연수이다. An electronic device according to an embodiment of the present disclosure includes a systolic array including a processor and a plurality of processing elements (PEs) arranged in a row and column direction, wherein the processor maps each weight component of a weight matrix to a corresponding row of the systolic array, sequentially inputs an input column vector of an input matrix in the row direction of the systolic array, and when a next input column vector is input to the systolic array after a preset number of input column vectors have been input to the systolic array, transfers an input component of the input column vector input to each column of M rows of the systolic array to a column next to each column of M+K rows, and outputs a cumulative sum produced through an operation of the weight matrix and the input matrix in response to the input column vectors being input to all rows of the systolic array, wherein M and K are preset natural numbers.","assignee":"삼성전자주식회사; 한양대학교 산학협력단","inventors":["최정욱","윤민용","김예은","소진인","김도원"],"publication_date":"2025-12-23","filing_date":"2025-12-04","priority_date":"2025-12-04","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G06N3/065","G","G06","G06F","G06F17/00","G06F17/10","G06F17/15","G06F17/153","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/544","G06F7/5443","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250177392A/en"},{"publication_number":"CN121190279A","title":"Intelligent Teaching System and Method Based on Large Language Model","abstract":"The invention relates to the technical field of intelligent teaching and discloses an intelligent teaching system and method based on a large language model. The system comprises five modules of teaching intention analysis, knowledge map adaptation, dynamic reasoning engine, teaching strategy generation and feedback optimization. The teaching intention analysis module disassembles the teaching interaction instruction into a knowledge domain label and a teaching behavior sequence through the semantic segmentation engine, and the knowledge map adaptation module matches the subject knowledge map according to the knowledge domain label and the teaching behavior sequence to extract an associated knowledge node set. The dynamic reasoning engine module inputs the node set into the large language model to complete multi-hop reasoning and generate an intermediate state vector containing a reasoning path. The teaching strategy generation module combines the vector path weights and the behavior sequence time stamps to generate a layered teaching strategy. The feedback optimization module collects user behavior data, updates knowledge graph matching rules through incremental learning, and adapts to diversified teaching requirements.","assignee":"Fujian Buke Information Technology Co ltd","inventors":["张林峰"],"publication_date":"2025-12-23","filing_date":"2025-11-27","priority_date":"2025-11-27","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/20","G06Q50/205","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121190279A/en"},{"publication_number":"CN121189294A","title":"Method and apparatus for generating response text","abstract":"本申请公开了一种答复文本生成方法及装置，涉及模型优化技术领域，包括：在模型训练过程中对特征训练矩阵进行量化张量列分解，以与权重矩阵和偏置进行计算；在模型推理的预填充阶段，利用模型训练后的张量分解形式的权重矩阵和偏置，与进行矩阵张量分解操作后的特征矩阵进行矩阵乘加运算；在解码阶段，对一维特征向量进行张量分解，以与张量分解形式的权重矩阵和偏置进行矩阵乘加，以基于对应的计算结果生成目标答复文本，解决了相关技术中，对矩阵分解计算过程中秩的选择要求较高，使得复杂度难以有效降低，或易导致利用分解形式得到的计算结果与原结果的误差较大的技术问题，达到了大幅降低存储和计算复杂度，有效提升推理性能的技术效果。 This application discloses a method and apparatus for generating response text, relating to the field of model optimization technology. The method includes: performing quantized tensor column decomposition on the feature training matrix during model training to calculate with the weight matrix and bias; in the pre-filling stage of model inference, using the weight matrix and bias in tensor decomposition form after model training, performing matrix multiplication and addition with the feature matrix after tensor decomposition; and in the decoding stage, performing tensor decomposition on the one-dimensional feature vector to perform matrix multiplication and addition with the weight matrix and bias in tensor decomposition form, thereby generating the target response text based on the corresponding calculation results. This solves the technical problem in related technologies where the rank selection requirement during matrix decomposition calculation is high, making it difficult to effectively reduce complexity, or easily leading to a large error between the calculation results obtained using the decomposition form and the original results. This method achieves the technical effect of significantly reducing storage and computational complexity and effectively improving inference performance.","assignee":"Suzhou Metabrain Intelligent Technology Co Ltd","inventors":["侯鹏宇","王彦伟"],"publication_date":"2025-12-23","filing_date":"2025-11-26","priority_date":"2025-11-26","cpc_codes":["G","G06","G06F","G06F40/00","G06F40/10","G06F40/166","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2411","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N5/00","G06N5/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121189294A/en"},{"publication_number":"CN121190892A","title":"A method and system for extracting elements from mountain pastures","abstract":"The application relates to the technical field of remote sensing information processing and geographic information systems, discloses a mountain pasture element extraction method and a mountain pasture element extraction system, and aims to solve the problems of insufficient local detail perception and multiscale information fusion capability of the existing remote sensing image classification. The method comprises the steps of preprocessing a remote sensing image, extracting and converting the remote sensing image into multi-dimensional spectrum-space joint characteristics of a multi-channel two-dimensional image, constructing and training a modified Swin-UNet model integrating a local perception enhancement module and a cross-layer characteristic fusion module, classifying mountain pasture elements by using the model, and dynamically evaluating the balance of grasses and livestock. By adopting the technical scheme, the application can remarkably improve classification precision, spatial continuity and robustness, and effectively supports dynamic evaluation and scientific management of the balance of the grass and livestock.","assignee":"Aba Natural Resources And Science And Technology Information Research Institute","inventors":["张晓锋","熊海霞","左世祥","余娟","张倩","陈建华","郑自强","王炳乾","张洪吉","宋馨平","杨晨成","陈青松","马硕","颜霜霜"],"publication_date":"2025-12-23","filing_date":"2025-11-26","priority_date":"2025-11-26","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06V","G06V10/00","G06V10/40","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/10","G06V20/188"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121190892A/en"},{"publication_number":"CN121188488A","title":"A model training method and system for electricity consumption information collection systems","abstract":"The application provides a model training method and a system applied to an electricity consumption information acquisition system, in the application, original electricity consumption data is always kept at a terminal, only model parameter update which cannot reversely push the original data is exchanged with an electric power main station, the privacy leakage problem is fundamentally solved, the model of each acquisition terminal is optimized through the local data, a specific load curve of the acquisition terminal can be better fitted, the prediction accuracy is obviously improved, meanwhile, the updating amount of the transmission model parameter is far smaller than the transmission original data, precious communication bandwidth is saved, and the model can continuously evolve along with the change of an electricity consumption mode of a user and has long-term adaptability.","assignee":"NANJING NENGRUI AUTOMATION EQUIPMENT CO Ltd","inventors":["仲敢","丁乔乐","许璐","杨健标","蒋爱","陈佳琪","周涛","吴蒙坤"],"publication_date":"2025-12-23","filing_date":"2025-11-25","priority_date":"2025-11-25","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F21/00","G06F21/60","G06F21/602","G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6245","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121188488A/en"},{"publication_number":"CN121188441A","title":"A Machine Learning-Based Method for Peak Flow Inversion","abstract":"本发明公开了一种基于机器学习的洪峰流量反演方法，涉及洪峰流量数据处理技术领域。该基于机器学习的洪峰流量反演方法，包括以下步骤：水文信号干扰监测；水文信号传输监测；水文信号数据质量监测。本发明通过水文信号干扰特征分析以判断是否采取动态抗干扰策略，其次进行洪峰水文信号传输完整性校验以判断是否采取传输链路优化，最后进行水文数据质量评估以判断是否进行洪峰反演，达到了提升洪峰流量反演结果准确性的效果，解决了现有技术中存在洪峰流量反演结果准确性低的问题。 This invention discloses a machine learning-based method for inverting peak flood discharge, relating to the field of peak flood discharge data processing technology. The machine learning-based method includes the following steps: hydrological signal interference monitoring; hydrological signal transmission monitoring; and hydrological signal data quality monitoring. This invention analyzes hydrological signal interference characteristics to determine whether a dynamic anti-interference strategy should be adopted; secondly, it verifies the integrity of peak flood hydrological signal transmission to determine whether transmission link optimization should be implemented; and finally, it assesses hydrological data quality to determine whether peak flood discharge inversion should be performed. This achieves the effect of improving the accuracy of peak flood discharge inversion results and solves the problem of low accuracy in existing technologies.","assignee":"PowerChina Chengdu Engineering Co Ltd","inventors":["余春涛","王涛","李铭","万民","任逍迪","王维"],"publication_date":"2025-12-23","filing_date":"2025-11-25","priority_date":"2025-11-25","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/217","G","G01","G01C","G01C13/00","G","G01","G01F","G01F1/00","G01F1/002","G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/211","G06F18/2113","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2131","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2431","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/24765","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N5/00","G06N5/01","H","H04","H04L","H04L63/00","H04L63/12","H04L63/123","Y","Y02","Y02A","Y02A10/00","Y02A10/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121188441A/en"},{"publication_number":"CN121188170A","title":"A content generation and interactive dialogue method and system based on AI.","abstract":"The invention discloses a content generation and interactive dialogue method and system based on AI artificial intelligence, which are characterized in that semantic tag sequences in multiple rounds of conversations are modeled, semantic derailment points are identified in a behavior evolution track template, and an intention clarification mechanism is triggered; the method comprises the steps of constructing a semantic tension map based on user feedback, forming a tension closed cluster and generating multipath response candidate fragments, utilizing a false guide chain cross analysis unit to identify a high risk response chain associated with historical negative feedback, eliminating shielding content, selecting an optimal response fragment through a context fracture span model and dynamically updating a behavior evolution model. The invention can obviously improve the semantic stability, consistency and safety of the dialogue system, reduce semantic drift, misleading answers and illusion risks, and has higher reliability and technical advancement. The system scheme can realize the modularized floor of the method function, and is suitable for being applied to the scenes of intelligent customer service, general large models, decision support systems and the like.","assignee":"Shanghai Zhutong Information Technology Co ltd","inventors":["陈东","黄剑","胡伟","温善祥","任飞"],"publication_date":"2025-12-23","filing_date":"2025-11-25","priority_date":"2025-11-25","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3344","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3346","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3347","G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G06F16/353","G","G06","G06F","G06F16/00","G06F16/30","G06F16/36","G06F16/367","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121188170A/en"},{"publication_number":"CN121189200A","title":"Three-dimensional piping layout method for complex FLNG ship compartments","abstract":"The application relates to the technical field of pipelines, in particular to a three-dimensional pipeline layout method of a complex FLNG cabin, which comprises the following steps of S1, initializing basic parameters, S2, adopting an intelligent initial strategy based on barrier density to calculate an initial solution, taking the optimal solution in the initial solution as a current global optimal solution, S3, entering an iterative process, calculating an energy factor, if the energy factor is greater than 1, carrying out long-distance migration with the probability of 30%, carrying out hole digging with the probability of 70%, otherwise, carrying out foraging or predator avoidance with the equal probability, S4, carrying out elite reservation and neighborhood search strategies periodically, S5, executing an adaptive orthogonal constraint relaxation strategy and boundary limiting operation, S6, calculating and determining a final global optimal solution, and outputting a final layout result. The application improves the calculation efficiency and the convergence speed, attaches to the actual condition of engineering, and reduces the engineering cost on the basis of meeting orthogonalization feasibility preferentially.","assignee":"China University of Petroleum East China","inventors":["吴磊","李广鑫","刘超","徐郎君","孙玉海","傅强","肖文生","马一歌","许登攀","魏晓康","冯启航","梅江涛","高靖坤","韩军明","王俊伟"],"publication_date":"2025-12-23","filing_date":"2025-11-25","priority_date":"2025-11-25","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06F","G06F30/00","G06F30/10","G06F30/15","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06F","G06F2113/00","G06F2113/14"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121189200A/en"},{"publication_number":"CN121190536A","title":"A method for monitoring the status of control signals of a smart seedling planting machine","abstract":"本发明提出一种树苗智能种植机控制信号状态监测方法，涉及智能农业信号处理领域；该方法包括：采集不同作业状态下的主控与辅助信号，进行统一格式化、异常剔除与时序补全，构建控制信号数据集；将主控与辅助信号映射为图像，构建图像变换矩阵，提取主控边缘特征生成主控结构感知图；依据结构感知图执行耦合调制，生成伪辅助态图像并与辅助图像组进行结构差异分析，得到动态变形图；基于变形图执行仿射增强与空间配准形成配准图像对；结合边缘保持机制进行特征融合生成融合状态特征图；计算边缘梯度差异，引入结构一致性因子并建立联合优化因数，对误差进行平衡调节，完成控制信号状态监测模型训练，实现复杂作业环境下的精准识别与稳定监测。 This invention proposes a method for monitoring the control signal status of an intelligent seedling planting machine, relating to the field of intelligent agricultural signal processing. The method includes: collecting master and auxiliary signals under different operating conditions, performing unified formatting, anomaly removal, and temporal completion to construct a control signal dataset; mapping the master and auxiliary signals to images, constructing an image transformation matrix, and extracting master control edge features to generate a master control structure perception map; performing coupled modulation based on the structure perception map to generate a pseudo-auxiliary state image and performing structural difference analysis with the auxiliary image group to obtain a dynamic deformation map; performing affine enhancement and spatial registration based on the deformation map to form a registered image pair; combining an edge-preserving mechanism to perform feature fusion to generate a fused state feature map; calculating edge gradient differences, introducing a structural consistency factor, and establishing a joint optimization factor to balance and adjust errors, completing the training of the control signal status monitoring model, and achieving accurate identification and stable monitoring under complex operating environments.","assignee":"Shandong Engineering Vocational and Technical University","inventors":["南庆霞","许艳春","赵棣","韩龙"],"publication_date":"2025-12-23","filing_date":"2025-11-25","priority_date":"2025-11-25","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/30","G06T7/33","G06T7/337","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06T","G06T7/00","G06T7/10","G06T7/13","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121190536A/en"},{"publication_number":"KR102901883B1","title":"Multilayer weight context integration system and method for generative artificial intelligence models","abstract":"본 개시는 사용자의 클라이언트 단말기에서 입력된 자연어 기반 질의에 응답하여 코드 생성을 수행하는 지능형 에이전트 시스템에 있어서, 상기 시스템은 클라이언트 단말기 및 인공지능 서버를 포함하고, 상기 인공지능 서버는, 상기 클라이언트 단말기로부터 수신된 사용자 입력 및 사용자 상태정보를 포함하는 단기 기억(STM) 정보를 이용하여 1차 컨텍스트를 생성하는 STM 모듈, 현재 작업(Task) 이력, 커밋 기록, 기능별 변경 내역, 빌드 및 테스트 로그를 포함하는 중기 기억(MTM) 정보 및 상기 1차 컨텍스트를 이용하여 2차 컨텍스트를 생성하는 MTM 모듈, 프로젝트 관련 장기 정보가 저장된 그래프 데이터베이스를 기반으로 요구사항, 사용예, 설계 명세, 구현 코드 및 과거 테스트 이력을 포함하는 장기 기억 정보(LTM) 및 상기 2차 컨텍스트를 이용하여 3차 컨텍스트를 생성하는 장기 기억(LTM) 모듈, 상기 단기 기억 정보, 중기 기억 정보 및 장기 기억 정보에 기초하여, 관련성, 최신성, 계층적 정보, 제약 조건 및 가이드라인, 정보 요약 및 필터링 및 충돌 해결을 포함하는 주요 룰에 기초하여 최종 컨텍스트를 생성하는 컨텍스트 생성 모듈 및 상기 최종 컨텍스트를 기반으로 사용자 요청에 따른 코드 생성, 질문 응답 및 작업 추천 중 적어도 하나의 동작을 수행하는 지능형 에이전트를 포함하는, 지능형 에이전트 시스템을 포함할 수 있다. The present disclosure relates to an intelligent agent system that performs code generation in response to a natural language-based query input from a user's client terminal, the system including a client terminal and an artificial intelligence server, wherein the artificial intelligence server comprises: an STM module that generates a primary context using short-term memory (STM) information including user input and user status information received from the client terminal; an MTM module that generates a secondary context using medium-term memory (MTM) information including current task history, commit records, functional change history, build and test logs, and the first context; a long-term memory (LTM) module that generates a tertiary context using long-term memory information (LTM) including requirements, usage examples, design specifications, implementation codes, and past test history based on a graph database in which long-term information related to a project is stored, and the second context; a context generation module that generates a final context based on major rules including relevance, recency, hierarchical information, constraints and guidelines, information summarization and filtering, and conflict resolution based on the short-term memory information, medium-term memory information, and long-term memory information; and a code generation module that generates, questions are answered, and tasks are recommended based on a user request based on the final context. An intelligent agent system may include an intelligent agent that performs at least one action.","assignee":"주식회사 프렌티스","inventors":["윤성열"],"publication_date":"2025-12-22","filing_date":"2025-11-25","priority_date":"2025-08-04","cpc_codes":["G","G06","G06F","G06F8/00","G06F8/30","G06F8/33","G","G06","G06F","G06F11/00","G06F11/36","G06F11/3668","G06F11/3672","G06F11/3684","G","G06","G06F","G06F11/00","G06F11/36","G06F11/3698","G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G06F16/9024","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9032","G06F16/90332","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9038","G","G06","G06F","G06F8/00","G06F8/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102901883B1/en"},{"publication_number":"CO2025017116A2","title":"Nanoemulsion adjuvant compositions for human papillomavirus vaccines","abstract":"La presente divulgación proporciona, entre otras cosas, una composición de vacuna que incluye un adyuvante de nanoemulsión de escualeno (SNE) y partículas similares a virus del HPV (VLP) de al menos un tipo del papilomavirus humano (HPV) seleccionado del grupo que consiste en los tipos del HPV: 6, 11, 16, 18, 26, 31, 33, 35, 39, 45, 51, 52, 53, 55, 56, 58, 59, 66, 68, 73 y 82. This disclosure provides, among other things, a vaccine composition that includes a squalene nanoemulsion (SNE) adjuvant and HPV virus-like particles (VLPs) of at least one human papillomavirus (HPV) type selected from the group consisting of HPV types: 6, 11, 16, 18, 26, 31, 33, 35, 39, 45, 51, 52, 53, 55, 56, 58, 59, 66, 68, 73, and 82.","assignee":"Merck Sharp & Dohme Llc","inventors":["Patrick L Ahl","William J Smith","John Gaspar","Julie M Skinner","Randal J Soukup","Nicole Lea Sullivan"],"publication_date":"2025-12-19","filing_date":"2025-12-09","priority_date":"2023-06-09","cpc_codes":["A","A61","A61K","A61K39/00","A61K39/12","G","G06","G06V","G06V20/00","G06V20/10","G06V20/13","A","A61","A61K","A61K39/00","A61K39/39","A","A61","A61P","A61P31/00","A61P31/12","A61P31/20","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06V","G06V10/00","G06V10/40","G06V10/62","G","G06","G06V","G06V20/00","G06V20/50","G06V20/52","G06V20/54","A","A61","A61K","A61K39/00","A61K2039/51","A61K2039/525","A61K2039/5258","A","A61","A61K","A61K39/00","A61K2039/555","A61K2039/55511","A61K2039/55566","C","C12","C12N","C12N2710/00","C12N2710/00011","C12N2710/20011","C12N2710/20023","C","C12","C12N","C12N2710/00","C12N2710/00011","C12N2710/20011","C12N2710/20031","C","C12","C12N","C12N2710/00","C12N2710/00011","C12N2710/20011","C12N2710/20034","C","C12","C12N","C12N7/00","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CO","kind":"application","source_url":"https://patents.google.com/patent/CO2025017116A2/en"},{"publication_number":"CN121167655A","title":"Hydrologic forecasting method and system based on self-adaptive correction chain","abstract":"本发明提供基于自适应修正链的水文预报方法及系统，涉及数据处理领域，该方法包括：根据降雨偏差溯源‑动态修正链和降雨观测数据，生成第一水库入库流量预报结果；确定当前场景的多维度特征，根据历史相似样本匹配‑规律应用链和当前场景的多维度特征，生成第二水库入库流量预报结果；根据产汇流知识‑阈值约束链和下垫面条件、降雨空间分布及流域状态，生成第三水库入库流量预报结果；确定第一水库入库流量预报结果、第二水库入库流量预报结果和第三水库入库流量预报结果的动态权重，生成最终的水库入库流量预报结果，具有提高水库入库流量预报结果的准确度的优点。 This invention provides a hydrological forecasting method and system based on an adaptive correction chain, relating to the field of data processing. The method includes: generating a first reservoir inflow forecast based on rainfall deviation source tracing-dynamic correction chain and rainfall observation data; determining multi-dimensional features of the current scenario and generating a second reservoir inflow forecast based on historical similar sample matching-regular application chain and the multi-dimensional features of the current scenario; generating a third reservoir inflow forecast based on runoff generation and confluence knowledge-threshold constraint chain, underlying surface conditions, spatial distribution of rainfall, and watershed status; and determining the dynamic weights of the first, second, and third reservoir inflow forecasts to generate the final reservoir inflow forecast, which has the advantage of improving the accuracy of reservoir inflow forecasts.","assignee":"Guodian Dadu River Hydropower Development Co Ltd","inventors":["陈在妮","牟时宇","丁柳丹","彭增","张峰","许剑","谢颖","犹鸿森"],"publication_date":"2025-12-19","filing_date":"2025-11-24","priority_date":"2025-11-24","cpc_codes":["G","G01","G01W","G01W1/00","G01W1/14","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/254","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121167655A/en"},{"publication_number":"CN121169933A","title":"Main cable erection quality defect on-line monitoring method and system based on image analysis","abstract":"本发明提出了基于图像分析的主缆架设质量缺陷在线监测方法及系统，属于悬索桥施工质量控制技术领域，本发明通过图像采集设备采集对应的索股图像，进行预处理之后，利用YOLO深度学习模型对预处理图像中存在缺陷的图像进行缺陷目标定位，再利用霍夫变换确定存在缺陷的图像中的扭绞角度，利用区域生长算法确定存在缺陷的图像中的表面损伤面积，根据扭绞角度判断扭绞情况，根据表面损伤面积判断表面损伤情况，从而在索股架设过程中进行在线质量监测；解决现有技术中通过人工巡检的方式进行质量检查，存在缺陷识别误差大和效率低的技术问题。 This invention proposes an online monitoring method and system for main cable erection quality defects based on image analysis, belonging to the field of suspension bridge construction quality control technology. The invention acquires corresponding cable strand images using image acquisition equipment, performs preprocessing, and then uses a YOLO deep learning model to locate defect targets in the preprocessed images. Next, Hough transform is used to determine the twisting angle in the defective images, and a region growing algorithm is used to determine the surface damage area. The twisting angle and surface damage area determine the surface damage condition, thus enabling online quality monitoring during cable strand erection. This solves the technical problems of large defect identification errors and low efficiency in existing technologies that rely on manual inspection for quality checks.","assignee":"East China Construction Co Ltd Of Cccc Second Highway Engineering Co ltd","inventors":["冯耀宗","李维生","耿祥云","李旭东","朱得祥","鲁航郗","王一凡","陈辛","杜洪池","艾国清","苏洋","喻胜刚","先正权","王坤","张辉","郑珍根","杨恩意","侯一萌","苏广","吴奔超"],"publication_date":"2025-12-19","filing_date":"2025-11-24","priority_date":"2025-11-24","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T7/00","G06T7/60","G06T7/62","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121169933A/en"},{"publication_number":"CN121167650A","title":"Expressway vehicle-road cooperative data management method and system based on multi-mode perception","abstract":"The application provides a method and a system for managing highway vehicle-road cooperative data based on multi-mode perception. The method comprises the steps of obtaining a multi-mode vehicle-road cooperation standard data set through obtaining and preprocessing the multi-mode vehicle-road cooperation data set, obtaining a vehicle-road cooperation effective characteristic data set through data extraction, generating a vehicle-road cooperation three-dimensional target characteristic map through dynamic fusion, analyzing by combining with constructed twin model construction, obtaining road network state change prediction data in a preset time period, generating lane-level decision data by combining the vehicle-road cooperation three-dimensional target characteristic map, pushing the lane-level decision data to a corresponding vehicle end, and simultaneously executing block chain storage and data access verification on the vehicle-road cooperation effective characteristic data set.","assignee":"Guangzhou Turing It Co ltd","inventors":["林宇","张湘涛","宋英良"],"publication_date":"2025-12-19","filing_date":"2025-11-21","priority_date":"2025-11-21","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06F","G06F16/00","G06F16/20","G06F16/27","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G08","G08G","G08G1/00","G08G1/01","G08G1/0104","G08G1/0125","G08G1/0129"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121167650A/en"},{"publication_number":"AU2025271481A1","title":"High-Capacity Storage of Digital Information in DNA","abstract":"A method for storage of an item of information (210) is disclosed. The method comprises encoding bytes (720) in the item of information (210), and representing using a schema the encoded bytes by a DNA nucleotide to produce a DNA sequence (230). The DNA sequence (230) is broken into a plurality of overlapping DNA segments (240) and indexing information (250) added to the plurality of DNA segments. Finally, the plurality of DNA segments (240) is synthesized (790) and stored (795).","assignee":"Europaisches Laboratorium fuer Molekularbiologie EMBL","inventors":["John BIRNEY","Nick GOLDMAN"],"publication_date":"2025-12-18","filing_date":"2025-11-28","priority_date":"2012-06-01","cpc_codes":["G","G06","G06F","G06F12/00","G06F12/02","G06F12/0223","G06F12/023","B","B82","B82Y","B82Y10/00","G","G06","G06N","G06N3/00","G06N3/12","G06N3/123","G","G11","G11C","G11C13/00","G11C13/02","G","G16","G16B","G16B30/00","G","G16","G16B","G16B50/00","G16B50/40","G","G16","G16B","G16B50/00","G16B50/50","G","G06","G06F","G06F2212/00","G06F2212/10","G06F2212/1032"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025271481A1/en"},{"publication_number":"AU2025271379A1","title":"Chemical impacts on a leach stockpile","abstract":"A method may comprise determining an acid gap based on a difference between total acid given and total acid consumption and determining an acid gap percentage based on the acid gap divided by the total acid consumption. The method may also include determining remaining soluble metal based on: (a percentage of leachable minerals) * (1- the acid gap percentage); and further adjusting at least one of leaching operations or a leaching model based on the remaining soluble metal. The method also includes transmitting an activation signal to activate a leaching device, wherein the leaching device performs the percolating of the total acid given from the second lift to the first lift, in response to the activation signal activating the leaching device that distributes the total acid given.","assignee":"Freeport Minerals Corp","inventors":["Rosemary D. Blosser","Amelia Briggs","Cristian Caro","Kevin Cheng","Raquel Crossman","Travis Gaddie","Dana GEISLINGER","Steven Chad Richardson","Margaret Alden Tinsley"],"publication_date":"2025-12-18","filing_date":"2025-11-26","priority_date":"2022-06-27","cpc_codes":["C","C22","C22B","C22B3/00","C22B3/04","C","C22","C22B","C22B15/00","C22B15/0063","C22B15/0065","C22B15/0067","C","C22","C22B","C22B15/00","C22B15/0063","C22B15/0065","C22B15/0067","C22B15/0071","C","C22","C22B","C22B3/00","C22B3/04","C22B3/06","C","C22","C22B","C22B3/00","C22B3/04","C22B3/06","C22B3/08","G","G01","G01N","G01N33/00","G01N33/24","G","G01","G01N","G01N5/00","G","G06","G06N","G06N20/00","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/067","G","G06","G06Q","G06Q50/00","G06Q50/02","B","B03","B03B","B03B7/00","C","C22","C22B","C22B15/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q2220/00"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025271379A1/en"},{"publication_number":"AU2025271383A1","title":"Tool selection for feature map encoding vs regular video encoding","abstract":"27238641v1 TOOL SELECTION FOR FEATURE MAP ENCODING VS REGULAR VIDEO ENCODING An apparatus for generating first encoded data and second encoded data. The apparatus comprises a determining unit for determining whether the apparatus generates encoded data including encoded data of a feature map based on a neural network. The apparatus also comprise an encoding unit for generating the first encoded data using a plurality of functions for encoding video data, in a case where the apparatus generates the first encoded data in a form of encoded video data not including the encoded data of the feature map. The encoding unit generates the encoded data of the feature map using a first part of the plurality of functions but not using a second part of the plurality of functions, in a case where the apparatus generates the second encoded data including the encoded data of the feature map.","assignee":"Canon Inc","inventors":["Christopher James ROSEWARNE"],"publication_date":"2025-12-18","filing_date":"2025-11-26","priority_date":"2021-04-07","cpc_codes":["H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","G","G06","G06N","G06N3/00","G06N3/02","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/771","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/12","H04N19/122","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/124","H","H04","H04N","H04N19/00","H04N19/46","H","H04","H04N","H04N19/00","H04N19/50","H04N19/503","H04N19/51","H04N19/513","H","H04","H04N","H04N19/00","H04N19/70","H","H04","H04N","H04N19/00","H04N19/85"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025271383A1/en"},{"publication_number":"AU2025271311A1","title":"Sensor-agnostic mechanical machine fault identification","abstract":"A method for identifying a fault of at least one mechanical machine, the comprising causing a first plurality of sensors coupled to a corresponding first plurality of mechanical machines to acquire a first plurality of sets of signals emanating from said first plurality 5 of mechanical machines, said first plurality of mechanical machines sharing at least one characteristic, supplying at least said first plurality of sets of signals of said first plurality of mechanical machines to a pre-existing fault classifier previously trained to automatically identify faults of a second plurality of mechanical machines based on signals emanating therefrom and previously acquired by a second plurality of sensors, 10 said second plurality of sensors being of a different type than said first plurality of sensors, said second plurality of mechanical machines sharing said at least one characteristic, modifying said pre-existing fault classifier by employing transfer learning, based at least on said first plurality of sets of signals of said first plurality of mechanical machines, thereby providing a modified fault classifier, applying said modified fault classifier to at 15 least one additional set of signals acquired by at least one sensor of said first plurality of sensors and emanating from at least one given mechanical machine sharing said at least one characteristic, said modified fault classifier being configured to automatically identify at least one fault of said at least one given mechanical machine based on said at least one additional set of signals, and providing a human sensible output, by an output device, 20 including at least identification of said fault of said at least one given mechanical machine, at least one of a repair or maintenance operation being performed based on said human sensible output, said first plurality of sensors and said second plurality of sensors being operative to sense mutually different types of signals.","assignee":"Augury Systems Ltd","inventors":["Daniel BARSKY","Gal BEN-HAIM","Christopher BETHEL","Ori NEGRI","Gal SHAUL","Saar YOSKOVITZ"],"publication_date":"2025-12-18","filing_date":"2025-11-25","priority_date":"2019-09-03","cpc_codes":["G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0259","G05B23/0275","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0218","G05B23/0224","G05B23/024","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0259","G05B23/0275","G05B23/0281","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0259","G05B23/0283","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0208","G05B23/0213","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025271311A1/en"},{"publication_number":"AU2025271199A1","title":"Multi-model timeseries forecasting of set-level variables","abstract":"A method includes predicting a first future timeseries for a first variable using a first model, predicting future values for a plurality of second variables using one or more second models, wherein the first variable is a function of the second variables, generating a second future timeseries for the first variable as based on the future values for the plurality of second variables, and providing a composite timeseries forecast for the first variable by combining the first future timeseries and the second future timeseries.","assignee":"Expedia Inc","inventors":["Stephen Andrew MCDONALD","Spoorthi MEDASANI","Anirudh Kamalapuram Muralidhar","Andrew Charles REUBEN"],"publication_date":"2025-12-18","filing_date":"2025-11-24","priority_date":"2023-05-05","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/02","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0202","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0206","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/14"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025271199A1/en"},{"publication_number":"KR20250175312A","title":"Memory device","abstract":"메모리 장치는, 전하가 채워진 상태가 제1 값으로 정의되고, 전하가 비워진 상태가 제2 값으로 정의되는 복수의 메모리 셀들을 포함하는 메모리 셀 어레이; 및 에러 정정 코드 회로를 포함하고, 상기 에러 정정 코드 회로는 신경망 연산을 위한 부호 있는 정수형(integer)의 복수의 가중치들을 포함하는 가중치 데이터에서, 제1 값의 사인 비트(Sign bit)를 가지며 상대적으로 다수의 가중치들이 분포된 제1 구간에 포함되는 가중치들이 제2 값의 사인 비트를 갖도록 상기 복수의 가중치들을 변환하며, 상기 변환된 가중치들을 포함하는 변환 가중치 데이터에서, 상기 변환된 가중치들의 비트 레벨에 따라 정해지는 비트 그룹들에 서로 다른 강도의 에러 정정 코드들을 적용함으로써 제1 패리티를 생성하고, 상기 가중치 데이터에 제1 패리티를 부가함으로써 가중치 코드워드를 생성한다. A memory device comprises a memory cell array including a plurality of memory cells, in which a state of being charged is defined as a first value and a state of being emptied is defined as a second value; and an error correction code circuit, wherein the error correction code circuit converts a plurality of weights of signed integer type for neural network operation, in weight data including the plurality of weights, such that weights included in a first section in which a relatively large number of weights are distributed have a sign bit of a first value, and generates a first parity by applying error correction codes of different strengths to bit groups determined according to bit levels of the converted weights in the converted weight data including the converted weights, and generates a weight codeword by adding the first parity to the weight data.","assignee":"삼성전자주식회사; 고려대학교 산학협력단","inventors":["정성우","이재윤"],"publication_date":"2025-12-16","filing_date":"2025-11-27","priority_date":"2025-05-30","cpc_codes":["G","G11","G11C","G11C29/00","G11C29/04","G11C29/08","G11C29/12","G11C29/38","G11C29/42","G","G06","G06F","G06F11/00","G06F11/07","G06F11/08","G06F11/10","G06F11/1008","G06F11/1048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250175312A/en"},{"publication_number":"CN121147232A","title":"Defects in plastic product production lines: AI vision sorting system","abstract":"本发明涉及塑料制品生产质量检测领域，公开了塑料制品生产线的缺陷AI视觉分拣系统，包括：数据采集单元，利用工业相机阵列对生产线上的塑料产品进行拍摄，生成原始图像数据；对原始图像数据进行动态背景去除处理，得到主体图像数据；特征提取单元，对主体图像数据进行多通道色彩空间转换，得到特征图像数据，本发明通过AI视觉系统，利用工业相机和深度学习模型，实现高精度、自动化的缺陷检测和分拣，提升生产线的分拣速度与准确度，减少人工干预，系统采用先进的背景去除、噪声过滤和图像增强技术，确保只处理有效的前景信息，并提高后续缺陷识别的鲁棒性和准确性，尤其在复杂环境下表现突出。 This invention relates to the field of quality inspection in plastic product manufacturing, and discloses an AI visual sorting system for defects in plastic product production lines. The system includes: a data acquisition unit that uses an industrial camera array to photograph plastic products on the production line and generate raw image data; dynamic background removal processing of the raw image data to obtain main image data; and a feature extraction unit that performs multi-channel color space conversion on the main image data to obtain feature image data. This invention, through an AI vision system, utilizes industrial cameras and deep learning models to achieve high-precision, automated defect detection and sorting, improving the sorting speed and accuracy of the production line and reducing manual intervention. The system employs advanced background removal, noise filtering, and image enhancement technologies to ensure that only valid foreground information is processed, and to improve the robustness and accuracy of subsequent defect identification, especially in complex environments.","assignee":"Nantong Younaite Plastic New Material Co ltd","inventors":["江庆国","江全绪","江汨绅"],"publication_date":"2025-12-16","filing_date":"2025-11-20","priority_date":"2025-11-20","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121147232A/en"},{"publication_number":"CN121147726A","title":"A Smart Generative Image Detection Method Based on Multi-Granularity Artifact Feature Fusion","abstract":"本发明涉及生成图像检测领域，提供一种基于多粒度伪影特征融合的智能生成图像检测方法，包括：收集生成图像和真实图像作为基础数据集；提取所述基础数据集中图像的多粒度伪影特征；利用所述多粒度伪影特征训练多粒度伪影特征融合检测模型；利用训练好的多粒度伪影特征融合检测模型进行生成图像检测。本发明一方面提取涵盖局部‑全局、时域‑频域的多粒度伪影特征；另一方面将多粒度伪影特征作为ViT模型的嵌入序列输入，利用Transformer编码器直接建模多粒度伪影特征之间全局关系的能力，利用多头自注意力机制实现多粒度伪影特征的自动关联融合。本发明方法综合利用不同粒度伪影特征的检测优势，提高生成图像检测的泛化性和鲁棒性。 This invention relates to the field of generated image detection, providing an intelligent generated image detection method based on multi-granularity artifact feature fusion. The method includes: collecting generated images and real images as a base dataset; extracting multi-granularity artifact features from the images in the base dataset; training a multi-granularity artifact feature fusion detection model using the multi-granularity artifact features; and performing generated image detection using the trained multi-granularity artifact feature fusion detection model. This invention extracts multi-granularity artifact features covering local-global and time-frequency domains; it also uses these multi-granularity artifact features as the embedding sequence input to a ViT model, leveraging the Transformer encoder's ability to directly model the global relationships between multi-granularity artifact features, and utilizing a multi-head self-attention mechanism to achieve automatic association and fusion of multi-granularity artifact features. This method comprehensively utilizes the detection advantages of different granularity artifact features, improving the generalization and robustness of generated image detection.","assignee":"CETC 30 Research Institute","inventors":["杨慧","康荣保","饶志宏","刘方","张志勇","陈剑锋","程丽君"],"publication_date":"2025-12-16","filing_date":"2025-11-20","priority_date":"2025-11-20","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/95","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121147726A/en"},{"publication_number":"CN121147893A","title":"A method, system, equipment and storage medium for detecting coal gangue","abstract":"本发明提供了一种煤矸石检测方法、系统、设备及存储介质，属于矿物加工处理领域，包括获取煤矸石图像；对煤矸石图像进行特征提取，得到多尺度特征图；对多尺度特征图进行加权计算及图像融合，得到增强的特征图；在增强的特征图中进行预测，通过损失函数定位边框损失，得到煤矸石的位置坐标和置信度。对煤矸石目标具有更好的适应性，提高了检测精度和速度，便于提高后续工业生产效率。 This invention provides a method, system, equipment, and storage medium for detecting coal gangue, belonging to the field of mineral processing. The method includes acquiring coal gangue images; extracting features from the coal gangue images to obtain multi-scale feature maps; performing weighted calculations and image fusion on the multi-scale feature maps to obtain enhanced feature maps; and performing prediction on the enhanced feature maps, using a loss function to locate bounding boxes, to obtain the position coordinates and confidence level of the coal gangue. This method has better adaptability to coal gangue targets, improves detection accuracy and speed, and facilitates increased efficiency in subsequent industrial production.","assignee":"China University of Mining and Technology Beijing CUMTB","inventors":["王洪梅","张伟龙","李世银","王保阳","李宗艳"],"publication_date":"2025-12-16","filing_date":"2025-11-19","priority_date":"2025-11-19","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/60","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06T","G06T7/00","G06T7/70","G06T7/73","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/40","G06V10/54","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121147893A/en"},{"publication_number":"CN121142011A","title":"A Soil Moisture Detection Method with Deep Learning Model for Optimized Accuracy","abstract":"本发明属于土壤水分检测技术领域，本发明提供了一种结合深度学习模型优化精度的土壤水分检测方法，包括：在多个土壤水分检测周期内，对每份土壤样本所划分的土样检测区域进行垂直方向上的水分检测，得到水分检测报告，对水分检测报告内每个土样检测区域对应的土样水分检测量进行误判分析，得到土样误检区域，从而有助于对检测结果进行误差分析。通过对比不同检测区域、不同深度以及不同周期的水分检测数据，能够找出可能存在的误差来源，确定土样误检区域，在深度学习模型训练过程中能够针对性地对土样误检区域进行重点训练，使模型更加关注如何准确检测这些容易误判的区域，从而优化模型在这些特定区域的检测性能，提高整体检测精度。 This invention belongs to the field of soil moisture detection technology. It provides a soil moisture detection method that combines deep learning models to optimize accuracy. The method includes: performing vertical moisture detection on the soil sample detection areas of each soil sample within multiple soil moisture detection cycles, obtaining a moisture detection report, and performing misjudgment analysis on the soil moisture detection amount corresponding to each soil sample detection area within the moisture detection report to identify false detection areas. This facilitates error analysis of the detection results. By comparing moisture detection data from different detection areas, depths, and cycles, potential error sources can be identified, and false detection areas can be determined. During the deep learning model training process, targeted training can be conducted on these false detection areas, making the model more focused on accurately detecting these easily misjudged areas, thereby optimizing the model's detection performance in these specific areas and improving overall detection accuracy.","assignee":"Northwest A&F University","inventors":["蔡耀辉","赵西宁","高晓东","孙世坤","刘柯楠","文明宜","吴亦霖"],"publication_date":"2025-12-16","filing_date":"2025-11-19","priority_date":"2025-11-19","cpc_codes":["G","G01","G01N","G01N33/00","G01N33/24","G01N33/246","G","G06","G06N","G06N20/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121142011A/en"},{"publication_number":"CN121146236A","title":"A method and system for intelligent dynamic scheduling of freight vehicles","abstract":"The invention relates to the technical field of logistics, in particular to an intelligent dynamic dispatching method and system for a freight vehicle, comprising the steps of determining a risk index of each road section unit according to the change condition of connected road section units, updating the risk index by combining the connection condition of the road section units and nodes, obtaining an updated risk index of each road section unit, adjusting the updated risk index by combining the road section units in a basic road network model at each moment according to the running speed of a current freight driver at each moment in each historical transportation process and the emergency braking quantity in each preset time period, obtaining a final risk index of each road section unit by combining the road section units at each moment, thus constructing a weighted road network model, combining a graph neural network model, and outputting the optimal path of the current freight driver from a starting point to an end point. The invention ensures the dispatching result to be safer by determining the optimal path.","assignee":"Shaanxi Yanchang Petroleum Dodge Logistics Technology Co ltd","inventors":["顾鑫","张文强","郭艳雯","支文博","薛虎军","刘林锋"],"publication_date":"2025-12-16","filing_date":"2025-11-19","priority_date":"2025-11-19","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/04","G06Q10/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/067","G","G06","G06Q","G06Q50/00","G06Q50/40","Y","Y02","Y02T","Y02T10/00","Y02T10/10","Y02T10/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121146236A/en"},{"publication_number":"CN121144473A","title":"Human-computer dialogue methods, software products, devices, and media based on large language models","abstract":"The application provides a man-machine conversation method based on a large language model, which comprises the steps of starting conversation from a starting node of a predefined conversation process, sending initial problem information configured by the starting node to a user, receiving initial reply information input by the user aiming at the initial problem information, determining a first intention recognition node based on the initial reply information, displaying a node problem corresponding to a current node to the user, receiving intention reply information input by the user aiming at the node problem, determining a subsequent node through the large language model by the intention reply information and a candidate option set predefined by the current node until the determined node is a reply node, and displaying reply information corresponding to the reply node to the user. The combination of the graphic flow and the large language model enables the dialogue path to be pushed according to preset logic, reduces the risk of deviating from a core target, and simultaneously enhances the naturalness and flexibility of the dialogue by utilizing the semantic understanding capability of the large language model, so that the user can better understand the reply.","assignee":"Shanghai Golden Bridge Information Technology Co ltd","inventors":["白传旭","冯健","周彦中","高海江","卞佳炜"],"publication_date":"2025-12-16","filing_date":"2025-11-19","priority_date":"2025-11-19","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3344","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3346","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3347","G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G06F16/353","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121144473A/en"},{"publication_number":"CN121145968A","title":"A large model compression method based on layer pruning and parameter sharing","abstract":"The invention discloses a large model compression method based on layer pruning and parameter sharing, belongs to the technical field of artificial intelligence and natural language processing, and solves the problems that memory occupation is high, calculation cost is high, and a traditional pruning method is not modeling interlayer dynamic relevance and parameter sharing technology is difficult to adapt to a post-pruning structure when a large language model is deployed. The technical scheme includes that relative transformation intensity between layers is calculated through forward propagation, dynamic importance indexes are obtained through combination of weight sparsity to perform layer pruning, a reserved layer is grouped based on cosine similarity, a father block is reserved with weights, sub blocks share weights and can learn scaling factors are introduced, a compressed model is subjected to light fine adjustment, and parameters of the father block, the scaling factors, a normalization layer and an output layer are only thawed. The invention can obviously reduce the volume of the model, improve the deployment efficiency and simultaneously maintain the performance of the model.","assignee":"Zhejiang University Of Technology Artificial Intelligence Innovation Research Institute Binjiang District Hangzhou City","inventors":["俞山青","谢文斌","陆耀","聂佳琦","赵尚上","钟高伟"],"publication_date":"2025-12-16","filing_date":"2025-11-19","priority_date":"2025-11-19","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121145968A/en"},{"publication_number":"CN121147225A","title":"Defect image generation method, apparatus, computer equipment and storage medium","abstract":"The application relates to a defect image generation method, a defect image generation device, computer equipment and a storage medium. The method comprises the steps of obtaining a target image and target defect guiding information, preprocessing the target image to obtain first image information, inputting the first image information and the target defect guiding information into a trained defect image generation model, outputting the target defect image, wherein the defect image generation model is used for obtaining a first target defect image when the input target defect guiding information comprises the target type defect guiding information, and obtaining a second target defect image when the input target defect guiding information comprises the target type defect guiding information and the target image defect guiding information, and the second target defect image is higher than the first target defect image in image accuracy. By adopting the application, the output with different image precision can be obtained according to different inputs.","assignee":"Shenzhen Smartmore Technology Co Ltd","inventors":["巫文良","陈鹏光","刘枢"],"publication_date":"2025-12-16","filing_date":"2025-11-18","priority_date":"2025-11-18","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","Y","Y02","Y02P","Y02P90/00","Y02P90/30"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121147225A/en"},{"publication_number":"KR20250174857A","title":"Smart Glasses for Bidirectional Sign Language and Speech Translation","abstract":"본 발명은, 착용자의 안면에 착용되는 수어 번역기 안경으로서, 렌즈 및 상기 렌즈를 지지하는 안경테와, 상기 안경테에서 연장되는 안경다리를 포함하는 안경 본체와, 착용 시 상기 착용자의 전방을 향하도록 상기 안경테에 배치되어, 상기 착용자의 정면에 위치한 상대방의 동작을 이미지 또는 영상 정보로 취득하는 카메라센서와, 상기 착용자가 발화하는 음성을 감지하여 음성 신호를 출력하는 음성인식센서와, 상기 수어 번역기 안경과 무선으로 연결되는 사용자 단말기 및 이어폰 과의 사이에서 데이터 송수신을 수행하는 무선 통신부와, 상기 카메라센서로부터 입력되는 이미지/영상 정보에 기초하여, 상기 상대방의 수어 동작을 인식하고, 상기 수어 동작에 대응하는 의미 정보를 텍스트 및/또는 음성 정보로 변환하여 상기 사용자 단말기의 디스플레이 또는 상기 이어폰을 통하여 출력하도록 제어하는 제어부와, 상기 음성인식센서로부터 입력되는 상기 착용자의 발화 음성에 기초하여, 상기 발화 음성을 텍스트 정보로 변환하고, 상기 텍스트 정보에 대응하는 수어 동작을 나타내는 수어 영상 또는 3차원 아바타 영상을 생성하며, 상기 수어 영상 또는 3차원 아바타 영상을 상기 사용자 단말기의 디스플레이 상에 실시간으로 재생하도록 제어하는 상기 제어부를 포함하는 인공지능 수어 번역기 안경에 관한 것이다. The present invention relates to sign language translator glasses worn on the face of a wearer, The present invention relates to artificial intelligence sign language translator glasses, comprising: a glasses body including a lens and a glasses frame that supports the lens, and glasses temples extending from the glasses frame; a camera sensor that is positioned on the glasses frame so as to face the front of the wearer when worn and acquires the motion of the other party located in front of the wearer as image or video information; a voice recognition sensor that detects a voice spoken by the wearer and outputs a voice signal; a wireless communication unit that performs data transmission and reception between the sign language translator glasses and a user terminal and earphones that are wirelessly connected; a control unit that recognizes the sign language motion of the other party based on image/video information input from the camera sensor, converts semantic information corresponding to the sign language motion into text and/or voice information, and controls the control unit to output the text and/or voice information through a display of the user terminal or the earphones; and a control unit that converts the spoken voice of the wearer input from the voice recognition sensor into text information, generates a sign language image or a 3D avatar image representing the sign language motion corresponding to the text information, and controls the control unit to play the sign language image or the 3D avatar image in real time on the display of the user terminal.","assignee":"강선우","inventors":["강선우"],"publication_date":"2025-12-15","filing_date":"2025-11-26","priority_date":"2025-11-26","cpc_codes":["G","G09","G09B","G09B21/00","G09B21/009","G","G06","G06F","G06F3/00","G06F3/01","G06F3/017","G","G06","G06N","G06N20/00","G","G06","G06T","G06T13/00","G06T13/20","G06T13/40","G","G06","G06V","G06V20/00","G06V20/50","G06V20/56","G06V20/58","G","G10","G10L","G10L15/00","G10L15/26","G","G10","G10L","G10L21/00","G10L21/06","G10L21/10","G","G10","G10L","G10L25/00","H","H04","H04W","H04W4/00","H04W4/80"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250174857A/en"},{"publication_number":"KR20250174551A","title":"Apparatus for Learning Calculating the Optimal Management Route for Power Transmission Facilities","abstract":"본 발명의 송전설비에 대한 최적 관리 경로 산정 장치는, 송전 설비 단위로 각 설비 중심의 특정 영역의 지도를 취득하는 지도 정보 취득부; 취득한 지도 상에서 중심이 되는 설비에 대하여 주변 진입로의 시점 후보 지역들을 선정하는 시점 후보 지역 선정부; 상기 특정 영역의 지도에 특정 간격으로 그어진 가상의 그리드의 교점을 각각의 노드로 구분하고, 상기 시점 후보 지역들과, 상기 그리드의 노드들 및 상기 설비의 중심점에 대하여, 길이, 경사로, 수목 밀도에 대한 정보들을 획득하여 반영하는 노드 구성부; 및 각 시점 후보 지역들로부터 상기 설비의 중심점까지의 최적 경로를 산정하는 최적 경로 도출부를 포함할 수 있다. The device for calculating an optimal management path for a power transmission facility of the present invention may include a map information acquisition unit that acquires a map of a specific area centered on each power transmission facility unit; a point candidate region selection unit that selects candidate start points of an access road surrounding a facility that is the center on the acquired map; a node configuration unit that divides the intersections of a virtual grid drawn at specific intervals on the map of the specific area into nodes, and acquires and reflects information on length, slope, and tree density for the point candidate regions, nodes of the grid, and the center point of the facility; and an optimal path derivation unit that calculates an optimal path from each of the candidate start points to the center point of the facility.","assignee":"한국전력공사","inventors":["권영운"],"publication_date":"2025-12-12","filing_date":"2025-11-25","priority_date":"2023-05-26","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/04","G06Q10/047","G","G01","G01C","G01C21/00","G01C21/26","G01C21/34","G01C21/3407","G","G01","G01C","G01C21/00","G01C21/26","G01C21/34","G01C21/3446","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/083","G06Q10/08355","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26","Y","Y04","Y04S","Y04S10/00","Y04S10/50"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250174551A/en"},{"publication_number":"CN121121129A","title":"A CT image segmentation and classification method for pulmonary embolism combined with quality assessment","abstract":"本发明公开了一种结合质量评价的CT图像肺栓塞分割分类方法，涉及图像处理的技术领域，其包括以下步骤：将256×256的肺栓塞CT图像及其质量评分输入质量评分引导编码器，通过线性变换扩展质量评分维度，与ResNet34提取的特征图逐层点乘融合，生成多尺度编码特征；通过小波变换融合跳跃链接模块对编码特征进行小波域分解与重构，优化特征传输；采用多尺度交叉增强解码器对特征进行多尺度反卷积融合，结合高效通道注意力机制输出分割结果，同时通过分类头输出肺栓塞存在性判断，该方法为肺栓塞的早期诊断、图像辅助诊断系统的开发以及临床应用提供了有力支持，具有广阔的应用前景和深远的社会意义。 This invention discloses a CT image pulmonary embolism segmentation and classification method combined with quality assessment, belonging to the field of image processing technology. The method includes the following steps: inputting a 256×256 pulmonary embolism CT image and its quality score into a quality score-guided encoder; expanding the quality score dimension through linear transformation; fusing it with feature maps extracted by ResNet34 layer by layer to generate multi-scale encoded features; performing wavelet domain decomposition and reconstruction of the encoded features through a wavelet transform fusion skip link module to optimize feature transmission; using a multi-scale cross-enhanced decoder to perform multi-scale deconvolution fusion of the features; and outputting the segmentation result by combining an efficient channel attention mechanism. Simultaneously, a classification head outputs a pulmonary embolism presence judgment. This method provides strong support for the early diagnosis of pulmonary embolism, the development of image-assisted diagnostic systems, and clinical applications, and has broad application prospects and profound social significance.","assignee":"Xuzhou Medical College","inventors":["冯祥浩","周奇","孙启航","侯乃龙","刘忠啸","丁涛","孙存杰","孟闫凯"],"publication_date":"2025-12-12","filing_date":"2025-11-17","priority_date":"2025-11-17","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/20","G06V10/26","G06V10/267","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/40","G06V10/52","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G06V10/765","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10072","G06T2207/10081","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30004","G06T2207/30061"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121121129A/en"},{"publication_number":"CN121116656A","title":"Expert model preloading methods, devices, chips, electronic devices, storage media, and computer program products","abstract":"本申请提供了一种专家模型预加载方法、装置、芯片、电子设备、存储介质及计算机程序产品，涉及数据处理领域，该方法包括：响应于专家模型被调用，获取多个专家模型的历史数据和模型参数，历史数据至少包括专家模型的被调用信息和性能信息，多个专家模型分别加载在显存、内存或硬盘中；根据专家模型的历史数据和模型参数确定专家模型的被调用预测值；根据被调用预测值对多个专家模型进行排序；将排序结果中前第一目标数量的专家模型加载至显存中，将排序结果中前第二目标数量的专家模型加载至内存中，将排序结果中前第三目标数量的专家模型加载至硬盘中。 This application provides an expert model preloading method, apparatus, chip, electronic device, storage medium, and computer program product, relating to the field of data processing. The method includes: in response to an expert model being invoked, acquiring historical data and model parameters of multiple expert models, wherein the historical data includes at least invocation information and performance information of the expert models, and the multiple expert models are respectively loaded into video memory, RAM, or hard disk; determining the invocation prediction value of the expert model based on the historical data and model parameters of the expert model; sorting the multiple expert models according to the invocation prediction value; loading the first target number of expert models in the sorting result into video memory, loading the second target number of expert models in the sorting result into RAM, and loading the third target number of expert models in the sorting result into hard disk.","assignee":"Shandong Yunhai Guochuang Cloud Computing Equipment Industry Innovation Center Co Ltd","inventors":["唐政","顾洪洋","曹成","王璞"],"publication_date":"2025-12-12","filing_date":"2025-11-17","priority_date":"2025-11-17","cpc_codes":["G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5011","G","G06","G06F","G06F11/00","G06F11/30","G06F11/3003","G06F11/3017","G","G06","G06F","G06F11/00","G06F11/30","G06F11/34","G06F11/3409","G06F11/3419","G06F11/3423","G","G06","G06F","G06F11/00","G06F11/30","G06F11/34","G06F11/3466","G06F11/3476","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5011","G06F9/5016","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121116656A/en"},{"publication_number":"CN121117525A","title":"A method and apparatus for time series prediction using large models for multi-level text alignment","abstract":"The invention provides a time sequence prediction method and a device for performing multistage text alignment by using a large model, and belongs to the technical field of time sequence prediction of a rail transit system based on the large model. The method comprises the steps of splitting a multi-variable time sequence input into a plurality of univariate time sequences according to characteristic dimensions, carrying out addition decomposition on each univariate time sequence, carrying out slicing treatment on each time sequence component after decomposition, embedding the slices into text embedding space of a pre-training language model, aligning the text embedding space with the text embedding space, combining the structured prompt with the aligned time sequence representation to form input of a large model, feeding the input combined with the prompt and the aligned representation into the frozen large language model, obtaining output representation of the model, and mapping the output representation into a final prediction result through a linear projection layer. According to the invention, the time series data and the natural language mode are effectively aligned and fused, so that the accuracy and the interpretability of the prediction are obviously improved.","assignee":"CRRC Changchun Railway Vehicles Co Ltd","inventors":["李波","高阳","麻竞文","周建航"],"publication_date":"2025-12-12","filing_date":"2025-11-17","priority_date":"2025-11-17","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121117525A/en"},{"publication_number":"CN121117194A","title":"An AI-based intelligent investment lead mining and precise matching method","abstract":"The invention provides an AI-based intelligent recruitment cue mining and accurate matching method. The method comprises the steps of collecting multisource recruitment data and preprocessing, adopting a heterogeneous embedding fusion mechanism to combine BERT and ELMO training intelligent recruitment clue mining models to generate fusion word vectors, carrying out clue classification and matching through a soft attention mechanism to identify potential client clues, collecting structured data of an enterprise side and a client side aiming at the potential client clues and preprocessing, constructing an intelligent recruitment accurate matching model based on time-enhanced space-time attention, introducing a gating network and a memory routing mechanism to carry out model training, outputting matching scores of potential clients and recruitment projects based on the trained accurate matching model, and determining matching results of the potential clients and the recruitment projects according to the matching scores. The invention obviously improves the conversion efficiency and the matching accuracy of the quotation cues.","assignee":"Shanghai Boran Zhongchuang Digital Technology Co ltd","inventors":["王伟","侯大波"],"publication_date":"2025-12-12","filing_date":"2025-11-17","priority_date":"2025-11-17","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3344","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/335","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121117194A/en"},{"publication_number":"CN121118266A","title":"A method for ship section layout based on critical polygons and dynamic microhabitats","abstract":"The invention belongs to the technical field of ship section two-dimensional irregular layout, and discloses a ship section layout method based on a critical polygon and a dynamic niche. The method comprises the steps of constructing an NFP solution method based on edge-vertex contact convex hulls to obtain critical polygons, constructing BL positioning stacking strategies to determine feasible domains based on the critical polygons to achieve accurate positioning of irregular polygons and calculate fitness, designing a multi-stage self-adaptive genetic algorithm parameter collaborative optimization method, dynamically adjusting genetic algorithm parameters based on population diversity and fitness variance, introducing a dynamic niche management and crowding punishment mechanism to avoid premature convergence of the population and maintain solution diversity, carrying out fine optimization on polygon rotation angles through genetic post-processing, and selecting a layout scheme preferentially. The method can effectively improve the space utilization rate of two-dimensional irregular layout, fully reserve the characteristics of the sectional complex shape in the stacking process, and is suitable for a ship building sectional stacking scene.","assignee":"Harbin Engineering University Sanya Nanhai Innovation And Development Base; Harbin Engineering University","inventors":["李敬花","赵佳庆","宋得宁","周磊","张博涵","金俊杰"],"publication_date":"2025-12-12","filing_date":"2025-11-17","priority_date":"2025-11-17","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/10","G06F30/15","B","B63","B63B","B63B71/00","B63B71/10","G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N3/00","G06N3/12","G06N3/126"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121118266A/en"},{"publication_number":"CN121120291A","title":"Intelligent cash management methods and systems for enterprises","abstract":"本申请提供了一种企业现金智能管理方法及系统，属于金融科技与人工智能技术领域。该方法包括：获取企业的历史现金流数据，构建数据的多维度时间特征空间，特征空间至少包括基于预设日历周期的第一类时间特征和基于企业特定业务周期的第二类时间特征；将多维度时间特征空间输入至预测模型，得到未来目标时段的现金流预测结果；基于现金流预测结果与历史现金流数据，动态校准基于米勒‑奥尔原理的现金管理模型中的最低控制线参数与现金流波动率参数；根据经动态校准后的现金管理模型，生成现金管理的决策依据。本申请解决了现有技术中现金管理模型参数静态僵化、无法自适应调整的问题，实现了管理策略的动态自适应，提升了决策的实时性与准确性。 This application provides a method and system for intelligent cash management for enterprises, belonging to the fields of financial technology and artificial intelligence. The method includes: acquiring historical cash flow data of an enterprise; constructing a multi-dimensional time feature space for the data, the feature space including at least a first type of time feature based on a preset calendar period and a second type of time feature based on a specific business cycle of the enterprise; inputting the multi-dimensional time feature space into a prediction model to obtain cash flow prediction results for future target periods; dynamically calibrating the minimum control line parameter and cash flow volatility parameter in a cash management model based on the Miller-Ol principle based on the cash flow prediction results and historical cash flow data; and generating a decision-making basis for cash management based on the dynamically calibrated cash management model. This application solves the problem of static and rigid cash management model parameters in existing technologies, which cannot be adaptively adjusted, and realizes dynamic adaptation of management strategies, improving the real-time performance and accuracy of decision-making.","assignee":"Steel Research International New Materials Innovation Center Shenzhen Co ltd","inventors":["李晗","李小群","辛梦婷","侯雅青","袁泽宇"],"publication_date":"2025-12-12","filing_date":"2025-11-17","priority_date":"2025-11-17","cpc_codes":["G","G06","G06Q","G06Q40/00","G06Q40/12","G06Q40/125","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0202","G","G06","G06F","G06F2123/00","G06F2123/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121120291A/en"},{"publication_number":"CN121114969A","title":"A method and system for processing airborne lidar depth sounding data","abstract":"The invention discloses a method and a system for processing airborne laser radar sounding data, which relate to the technical field of ocean surveying and earth observation and comprise the following steps of collecting original full waveform data of an airborne laser radar sounding system, wherein the original full waveform data comprises infrared laser channel data and blue-green laser multichannel data; and performing joint denoising processing on the original full-waveform data to obtain denoised waveform data, wherein the joint denoising processing comprises background noise modeling and removing based on machine learning and random noise filtering based on frequency domain low-pass filtering, and a collaborative joint denoising strategy is formed by fusing the background noise modeling based on machine learning and the frequency domain low-pass filtering based on signal-to-noise ratio optimization. The method not only can accurately estimate and remove complex background noise, but also can adaptively filter random noise, reserve weak underwater echo signals to the greatest extent, and effectively avoid the problem of water depth information loss caused by excessive smoothing.","assignee":"Geophysical Prospecting Surveying Team Shandong Bureau Of Coal Geology","inventors":["朱延华","黄士红","徐勇"],"publication_date":"2025-12-12","filing_date":"2025-11-17","priority_date":"2025-11-17","cpc_codes":["G","G01","G01S","G01S7/00","G01S7/48","G01S7/4802","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/23","G06F18/232","G","G06","G06F","G06F18/00","G06F18/20","G06F18/23","G06F18/232","G06F18/2321","G06F18/23213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2411","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/24317","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06F","G06F2123/00","G06F2123/02","G","G06","G06F","G06F2218/00","G06F2218/02","G06F2218/04","G","G06","G06F","G06F2218/00","G06F2218/12"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121114969A/en"},{"publication_number":"CN121116207A","title":"Data quantization circuits, data processing methods, processors, devices, and media based on lookup tables","abstract":"The present disclosure provides a look-up table based data quantization circuit, a data processing method, a processor, a device and a medium. The data quantization circuit comprises a data block preprocessing unit, an index alignment unit, an index generation unit and an output encoding unit, wherein the data block preprocessing unit is configured to determine target input data with the largest index part in an input data block, the index alignment unit is configured to align the index parts of a plurality of input data to the index values according to the index values of the index parts of the target input data to obtain a plurality of intermediate data, the index generation unit is configured to extract the mantissa parts of the plurality of intermediate data and combine the mantissa parts with N scaling factor mantissa part candidates to obtain N first indexes corresponding to each input data, and the output encoding unit is configured to access a lookup table according to N first indexes corresponding to each input data to determine target scaling factors and quantization results of the input data block. According to the method and the device, the optimal scaling factor is dynamically selected through the lookup table, so that higher quantization precision is ensured, the realization cost and delay are reduced, and the hardware efficiency is improved.","assignee":"Shanghai Bi Ren Technology Co ltd","inventors":["请求不公布姓名"],"publication_date":"2025-12-12","filing_date":"2025-11-17","priority_date":"2025-11-17","cpc_codes":["G","G06","G06F","G06F3/00","G06F3/06","G06F3/0601","G06F3/0602","G06F3/0608","G","G06","G06F","G06F3/00","G06F3/06","G06F3/0601","G06F3/0628","G06F3/0638","G06F3/064","G","G06","G06F","G06F3/00","G06F3/06","G06F3/0601","G06F3/0628","G06F3/0655","G06F3/0656","G","G06","G06F","G06F3/00","G06F3/06","G06F3/0601","G06F3/0668","G06F3/0671","G06F3/0673","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121116207A/en"},{"publication_number":"KR20250173994A","title":"ESG Strategy Digital Twin-based Multi-objective Portfolio Optimization System and Method","abstract":"본 발명은 기업의 ESG 전략을 디지털트윈 기반으로 모델링하고, 다목적 포트폴리오 최적화를 통해 ESG 성과·재무 성과·리스크를 동시에 고려하는 전략 포트폴리오를 설계·추천·학습하는 ESG 전략 디지털트윈 기반 다목적 포트폴리오 최적화 시스템 및 방법{ESG Strategy Digital Twin based Multi-objective Portfolio Optimization System and Method}에 관한 것이다. 본 발명에 따른 시스템은, ESG 전략별로 비용, 예상 ESG 효과, 예상 재무 효과, 리스크 변화, 수행 기간, 적용 조직, 선행조건, 의존관계, 태그 정보를 포함하는 전략 모델 데이터를 정형화하여 관리하는 전략 모델 관리 모듈과, ESG 지표·재무 지표·리스크 지표 및 예산·인력·설비 용량 등의 자원 제약을 통합한 기업 상태 벡터를 정의하고 전략 실행에 따라 상태를 갱신하는 기업 상태 디지털트윈 모듈을 포함한다. 또한 제약조건과 전략 후보군을 바탕으로 복수의 포트폴리오 시나리오를 생성하고, 각 시나리오에 대한 ESG·재무·리스크 지표의 시간 경과 효과를 시뮬레이션하는 시나리오 생성 및 시뮬레이션 모듈과, 상기 시뮬레이션 결과를 입력으로 하여 ESG 성과·재무 성과·리스크 감소를 독립된 목적함수로 정의하고 예산·인력·기간·규제·리스크 허용도 제약을 고려하는 다목적 최적화 연산을 수행하여 Pareto 최적 포트폴리오 집합을 산출하는 다목적 포트폴리오 최적화 모듈을 포함한다. 더 나아가, 본 발명은 Pareto 최적 포트폴리오 집합으로부터 사용자 선호 및 위험 성향을 반영한 복수의 대표 포트폴리오를 선택하고, 각 포트폴리오의 투자비, ESG 성과, 재무 성과, 리스크 프로파일을 시각화하여 의사결정자에게 제공하는 전략 추천 및 시각화 모듈과, 선택된 포트폴리오의 실제 실행 로그 및 성과 데이터를 수집하여 시뮬레이션 예측값과의 오차를 분석하고 그 결과를 전략 모델 데이터 및 디지털트윈 파라미터에 반영함으로써 시간이 지날수록 예측 정확도와 최적화 품질을 향상시키는 실행 결과 수집 및 학습 모듈을 포함한다. 따라서 본 발명에 따르면, 종래의 정태적 ESG 평가·보고 도구와 달리, ESG 전략을 전략 단위 데이터 자산으로 체계화하고, ESG·재무·리스크·자원 제약을 통합한 디지털트윈과 다목적 포트폴리오 최적화를 결합함으로써, 중·장기 관점의 ESG 전략 포트폴리오 설계·시뮬레이션·최적화·실행·학습이 하나의 폐루프 내에서 이루어지는 통합 ESG 전략 의사결정 플랫폼을 구현할 수 있다. The present invention relates to an ESG Strategy Digital Twin based Multi-objective Portfolio Optimization System and Method, which models a company's ESG strategy based on a digital twin and designs, recommends, and learns a strategic portfolio that simultaneously considers ESG performance, financial performance, and risk through multi-objective portfolio optimization. The system according to the present invention includes a strategy model management module that formats and manages strategy model data including costs, expected ESG effects, expected financial effects, risk changes, implementation periods, applicable organizations, prerequisites, dependencies, and tag information for each ESG strategy, and a corporate state digital twin module that defines a corporate state vector that integrates ESG indicators, financial indicators, risk indicators, and resource constraints such as budget, human resources, and facility capacity, and updates the state according to strategy execution. In addition, the system includes a scenario generation and simulation module that generates multiple portfolio scenarios based on constraints and strategic candidates and simulates the time-lapse effects of ESG, financial, and risk indicators for each scenario, and a multi-objective portfolio optimization module that defines ESG performance, financial performance, and risk reduction as independent objective functions using the simulation results as input and performs multi-objective optimization operations that consider budget, human resources, period, regulation, and risk tolerance constraints to produce a Pareto optimal portfolio set. Furthermore, the present invention includes a strategy recommendation and visualization module that selects multiple representative portfolios reflecting user preferences and risk propensity from a set of Pareto optimal portfolios, and provides the investment cost, ESG performance, financial performance, and risk profile of each portfolio to decision makers by visualizing them, and an execution result collection and learning module that collects actual execution logs and performance data of the selected portfolios, analyzes the error with the simulation prediction value, and reflects the results in the strategy model data and digital twin parameters, thereby improving prediction accuracy and optimization quality over time. Therefore, according to the present invention, unlike conventional static ESG evaluation and reporting tools, by systematizing ESG strategies as strategic unit data assets and combining digital twins that integrate ESG, finance, risk, and resource constraints with multi-objective portfolio optimization, an integrated ESG strategy decision-making platform can be implemented in which ESG strategy portfolio design, simulation, optimization, execution, and learning from a mid- to long-term perspective are performed within a single closed loop.","assignee":"박홍석","inventors":["박홍석"],"publication_date":"2025-12-11","filing_date":"2025-11-24","priority_date":"2025-11-24","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G","G06","G06N","G06N3/00","G06N3/12","G06N3/126","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06393","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/067","G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/105","G","G06","G06Q","G06Q40/00","G06Q40/06","G","G06","G06T","G06T19/00","G06T19/003","G","G06","G06F","G06F2123/00","G06F2123/02"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250173994A/en"},{"publication_number":"KR20250173997A","title":"Real-time face replacement system using deepfake technology","abstract":"본 발명은 영상 처리 기술 및 인공지능(AI) 기술 분야에 관한 것으로, 특히 딥페이크(Deepfake) 기술을 활용하여 영상 내 인물의 얼굴을 사용자가 설정한 기준(예: 연령, 성별, 표정 등)에 따라 실시간으로 변환하거나 가상의 얼굴로 대체함으로써 초상권 문제를 해결하고 사용자 맞춤형 푸티지 제작을 가능하게 하는 플랫폼에 관한 것이다. 또한 본 발명은 영상 콘텐츠 제작, 편집, 및 보호 기술 분야에서 활용될 수 있으며, 특히 초상권 관리가 중요한 미디어, 광고, 교육 및 엔터테인먼트 산업에서 유용하다. The present invention relates to the fields of image processing technology and artificial intelligence (AI) technology, and more particularly, to a platform that utilizes Deepfake technology to convert the face of a person in a video in real time or replace it with a virtual face according to criteria set by the user (e.g., age, gender, expression, etc.), thereby solving the portrait rights issue and enabling the production of user-customized footage. In addition, the present invention can be utilized in the fields of video content production, editing, and protection technology, and is particularly useful in the media, advertising, education, and entertainment industries where portrait rights management is important.","assignee":"손철희","inventors":["손철희"],"publication_date":"2025-12-11","filing_date":"2025-11-24","priority_date":"2024-11-25","cpc_codes":["G","G06","G06T","G06T19/00","G06T19/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06T","G06T11/00","G06T11/60","G","G06","G06T","G06T15/00","G06T15/04","G","G06","G06T","G06T15/00","G06T15/50","G","G06","G06T","G06T3/00","G06T3/10","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30196","G06T2207/30201"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250173997A/en"},{"publication_number":"KR20250173992A","title":"LEARNING METHOD OF TTS(Text-To-Speech) MODEL, TTS DEVICE AND METHOD OF PROVIDING TTS SERVICE USING TTS DEVICE","abstract":"본 발명은, 텍스트를 음성으로 변환하는 메인 블록과, 발화 특징을 추출하여 메인 블록에 반영하는 스피커 인코더를 포함하는 TTS(Text-To-Speech) 모델의 학습 방법에 있어서, 제1 대상의 제1 음성 데이터를 이용하여, 텍스트가 복수의 국가의 언어 중 적어도 하나의 음성으로 변환되도록 메인 블록을 학습시키고, 발화 특징을 제2 대상에 대하여 추출하여, 메인 블록을 프리즈하고, 제2 대상의 제2 음성 데이터를 이용하여 스피커 인코더를 학습시키는, TTS 모델의 학습 방법을 제공한다. The present invention provides a method for training a Text-To-Speech (TTS) model, which includes a main block that converts text into speech, and a speaker encoder that extracts speech features and reflects them to the main block, wherein the main block is trained using first speech data of a first subject so that text is converted into speech in at least one of a plurality of languages, speech features are extracted for a second subject, the main block is frozen, and the speaker encoder is trained using second speech data of the second subject.","assignee":"주식회사 포티투마루","inventors":["김동환","정우태","유성현","박주식"],"publication_date":"2025-12-11","filing_date":"2025-11-21","priority_date":"2023-12-20","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G10","G10L","G10L13/00","G10L13/02","G","G10","G10L","G10L13/00","G10L13/08","G","G10","G10L","G10L15/00","G","G10","G10L","G10L15/00","G10L15/005","G","G10","G10L","G10L15/00","G10L15/02","G","G10","G10L","G10L15/00","G10L15/04","G","G10","G10L","G10L15/00","G10L15/06","G","G10","G10L","G10L15/00","G10L15/06","G10L15/063","G","G10","G10L","G10L25/00","G10L25/27","G10L25/30"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250173992A/en"},{"publication_number":"KR20250173470A","title":"AI ESG Knowledge Engine-based Corporate Report Analysis, Evaluation and Journal Generation System","abstract":"본 발명은 AI 기반 ESG Knowledge Engine을 이용하여, 기업의 지속가능경영보고서 등 ESG 관련 비정형 비재무 보고서 문서(PDF, 스캔 이미지, HTML, 워드 파일 등)를 입력으로 받아 자동으로 분석·정규화·평가하고, 규제 적합성 및 리스크 수준을 산출한 후, 이를 기반으로 ESG 기사·저널을 자동 생성하는 시스템에 관한 것이다. 본 시스템은 기업 업로드 모듈, AI 문서 파싱 엔진, KPI 추출 및 정규화 엔진, 규제 매핑 엔진, ESG Scorecard 산출 엔진, AI 저널링 엔진으로 구성된다. 기업이 상기 비정형 ESG 보고서를 업로드하면, 시스템은 ESG KPI를 ESG 세그먼트 단위로 구조화하고, 산업·국가 기준에 따라 값과 단위를 정규화하며, CSRD/ESRS 등 규제 DB와 매핑하여 공시 수준과 미공시 항목을 평가하고 Compliance Score 및 미공시 항목 리스트를 산출한다. 최종적으로 ESG Scorecard와 핵심 인사이트가 포함된 기사 원고를 자동 생성함으로써, 분석 결과를 규제 대응, 의사결정 및 대외 커뮤니케이션에 바로 활용할 수 있게 한다. The present invention relates to a system that utilizes an AI-based ESG Knowledge Engine to automatically analyze, normalize, and evaluate ESG-related non-structured report documents (PDF, scanned images, HTML, Word files, etc.) such as corporate sustainability reports, calculates regulatory compliance and risk levels, and then automatically generates ESG articles and journals based on these. The system comprises a corporate upload module, an AI document parsing engine, a KPI extraction and normalization engine, a regulatory mapping engine, an ESG scorecard calculation engine, and an AI journaling engine. When a company uploads the above-mentioned non-structured ESG report, the system structures ESG KPIs into ESG segment units, normalizes values and units according to industry and national standards, and maps them to regulatory databases such as the CSRD/ESRS to evaluate the level of disclosure and undisclosed items, and calculates a Compliance Score and a list of undisclosed items. Ultimately, by automatically generating an article manuscript containing an ESG scorecard and key insights, the system enables the analysis results to be immediately utilized for regulatory response, decision-making, and external communication.","assignee":"박홍석","inventors":["박홍석"],"publication_date":"2025-12-10","filing_date":"2025-11-21","priority_date":"2025-11-21","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06393","G","G06","G06F","G06F16/00","G06F16/30","G06F16/34","G06F16/345","G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G","G06","G06Q","G06Q10/00","G06Q10/10","G","G06","G06V","G06V30/00","G06V30/10","G06V30/20","G","G06","G06V","G06V30/00","G06V30/40","G06V30/41","G06V30/412","G","G06","G06V","G06V30/00","G06V30/40","G06V30/41","G06V30/413"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250173470A/en"},{"publication_number":"KR20250172791A","title":"Method, apparatus, and program for generating artificial intelligence recommendation model","abstract":"사용자 데이터 기반 딥러닝을 통한 추천 편집점 정교화 방법, 서버 및 컴퓨터프로그램이 제공된다. 본 발명의 다양한 실시예에 따른 사용자 데이터 기반 딥러닝을 통한 추천 편집점 정교화 방법은, 컴퓨팅 장치에 의해 수행되는 방법에 있어서, 복수의 사용자 각각에 대응하는 복수의 스트리밍 컨텐츠 정보 및 복수의 편집 이력 정보를 획득하여 편집 스타일 데이터베이스르 구축하는 단계, 상기 편집 스타일 데이터베이스를 통해 학습 데이터 세트를 획득하는 단계 및 상기 학습 데이터 세트를 통해 하나 이상의 네트워크 함수에 대한 학습을 수행하여 상기 복수의 사용자 각각에 대응하는 복수의 맞춤 편집점 추천 모델을 생성하는 단계를 포함할 수 있다. A method, a server, and a computer program for refining recommended edit points through user data-based deep learning are provided. The method for refining recommended edit points through user data-based deep learning according to various embodiments of the present invention may include, in a method performed by a computing device, a step of acquiring a plurality of streaming content information and a plurality of editing history information corresponding to each of a plurality of users to construct an editing style database, a step of acquiring a learning data set through the editing style database, and a step of generating a plurality of customized editing point recommendation models corresponding to each of the plurality of users by performing learning on one or more network functions through the learning data set.","assignee":"주식회사 알피엠밸류","inventors":["이현우","추성훈","홍기용","최우진"],"publication_date":"2025-12-09","filing_date":"2025-12-04","priority_date":"2021-05-28","cpc_codes":["H","H04","H04N","H04N21/00","H04N21/40","H04N21/45","H04N21/466","H04N21/4662","H04N21/4666","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","H","H04","H04N","H04N21/00","H04N21/40","H04N21/43","H04N21/439","H04N21/4394","H","H04","H04N","H04N21/00","H04N21/40","H04N21/43","H04N21/44","H04N21/44008","H","H04","H04N","H04N21/00","H04N21/40","H04N21/47","H04N21/478","H04N21/4788","H","H04","H04N","H04N21/00","H04N21/80","H04N21/85","H04N21/854","H04N21/8549"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250172791A/en"},{"publication_number":"CO2025016747A2","title":"Low-complexity nn-based loop filter architectures with separable convolution","abstract":"RESUMEN Se describen técnicas de ejemplo para filtrar datos de video. Un dispositivo de ejemplo para al menos uno de codificar o decodificar datos de video incluye una o más memorias configuradas para almacenar los datos de video y uno o más procesadores. El uno o más procesadores están configurados para recibir una imagen de datos de video y reconstruir la imagen de los datos de video. El uno o más procesadores también están configurados para aplicar un filtro basado en red neuronal (NN) a la imagen reconstruida de datos de video. El filtro basado en NN incluye un filtro unificado. El filtro unificado incluye un bloque de cabecera, un bloque de transición, uno o más bloques troncales y un bloque de cola. Al menos uno del bloque de cabecera, el bloque de transición, el uno o más bloques troncales, o el bloque de cola incluye una descomposición poliádica canónica (CP) con convolución separable. ABSTRACT Example techniques for filtering video data are described. An example device for encoding or decoding video data includes one or more memories configured to store the video data and one or more processors. The one or more processors are configured to receive a video data image and reconstruct the video data image. The one or more processors are also configured to apply a neural network (NN)-based filter to the reconstructed video data image. The NN-based filter includes a unified filter. The unified filter includes a header block, a transition block, one or more trunk blocks, and a tail block. At least one of the header block, the transition block, the one or more trunk blocks, or the tail block includes a canonical polyadic (CP) decomposition with separable convolution.","assignee":"Qualcomm Inc","inventors":["Marta Karczewicz","Dmytro Rusanovskyy","Yun Li"],"publication_date":"2025-12-09","filing_date":"2025-12-01","priority_date":"2023-06-12","cpc_codes":["H","H04","H04N","H04N19/00","H04N19/42","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","H","H04","H04N","H04N19/00","H04N19/10","H04N19/169","H04N19/17","H04N19/176","H","H04","H04N","H04N19/00","H04N19/80","H04N19/82"],"country":"CO","kind":"application","source_url":"https://patents.google.com/patent/CO2025016747A2/en"},{"publication_number":"KR20250172512A","title":"Meat product sales method based on displaying flavor information and meat product sales method by recommending customer-customized meat product based on displaying flavor information","abstract":"본 발명은 소비자에게 판매하는 육제품이 가지는 풍미 정보(육향, 감칠맛, 연도, 육즙 등)를 표시하여 육제품을 판매하는 방법과 판매 육제품의 풍미 정보 표시에 기반하여 소비자의 취향에 부합하는 맞춤형 육제품을 추천하여 판매하는 방법에 관한 것이다. 이를 위해, 본 발명에 따른 풍미 정보 표시에 기반하여 육제품을 판매하는 방법은 육제품 판매 시스템이 소비자에게 판매하고자 하는 복수의 육제품들에 대하여, 각 육제품 별로 소비자가 취식할 때 지각할 수 있는 풍미(flavor) 정보를 데이터베이스에 미리 저장하는 육제품 정보 저장 단계; 육제품 판매 시스템이 사용자 단말을 통해 소비자에게 판매하기 위한 육제품에 대한 제품 정보를 제공하되, 제품 정보에 육제품에 대한 풍미 정보를 포함하여 표시하는 육제품 정보 제공 단계; 및 육제품 판매 시스템이 사용자 단말을 통해 육제품에 대한 풍미 정보를 확인한 소비자로부터 육제품 구매 요청을 입력받는 육제품 판매 단계를 포함한다. The present invention relates to a method for selling meat products by displaying flavor information (meat aroma, umami, tenderness, meat juice, etc.) of meat products sold to consumers, and a method for recommending and selling customized meat products that suit consumers' tastes based on the flavor information display of the meat products sold. To this end, the method for selling meat products based on the flavor information display according to the present invention comprises: a meat product information storage step in which a meat product sales system stores in advance, in a database, flavor information that consumers can perceive when eating each meat product among a plurality of meat products to be sold to consumers; a meat product information providing step in which the meat product sales system provides product information on meat products to be sold to consumers through a user terminal, wherein the product information includes and displays the flavor information of the meat product; and a meat product sales step in which the meat product sales system receives a meat product purchase request from a consumer who has confirmed the flavor information of the meat product through the user terminal.","assignee":"주식회사 인큐시스","inventors":["이진욱","허진석"],"publication_date":"2025-12-09","filing_date":"2025-11-26","priority_date":"2022-11-25","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0631","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0281","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0282","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0633"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250172512A/en"},{"publication_number":"KR20250172506A","title":"Neural network system and implementing method thereof","abstract":"뉴럴 네트워크 시스템에 있어서, 어레이 형태로 배열된 제1 메모리 셀을 포함하는 뉴럴 네트워크 회로; 및 상기 뉴럴 네트워크 회로의 로우(row) 라인 또는 컬럼(column) 라인에 전기적으로 연결되고, 복수의 타겟 메모리 셀이 기설정된 타겟 가중치를 갖도록 상기 연결된 로우 라인 또는 컬럼 라인에 전류를 인가하는 자기 참조 회로;를 포함하고, 상기 타겟 메모리 셀은, 상기 자기 참조 회로가 연결된 로우 라인 또는 컬럼 라인에 위치한 모든 메모리 셀인, 시스템을 제공할 수 있다. In a neural network system, a system can be provided, comprising: a neural network circuit including first memory cells arranged in an array form; and a self-reference circuit electrically connected to a row line or a column line of the neural network circuit and applying current to the connected row line or column line so that a plurality of target memory cells have preset target weights; wherein the target memory cells are all memory cells located in the row line or the column line to which the self-reference circuit is connected.","assignee":"주식회사 페블스퀘어","inventors":["이충현"],"publication_date":"2025-12-09","filing_date":"2025-11-24","priority_date":"2023-03-07","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G06N3/065","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G11","G11C","G11C11/00","G11C11/54"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250172506A/en"},{"publication_number":"KR20250172913A","title":"Insurance benefit payable item discovery and document recommendation system and method","abstract":"본 발명은 보험금 지급 가능 항목 발굴 및 서류 추천 시스템 및 방법에 관한 것이다. 보다 구체적으로는 사용자 단말에서 업로드된 진단서, 진료비 세부내역서, 수술확인서, 입·퇴원확인서 등의 청구 서류와, 하나 이상의 보험사 서버에 저장된 보험 계약 정보 및 약관 전문을 광학문자인식(OCR)과 자연어 처리(NLP) 기술을 이용하여 구조화하고, 약관에서 추출된 보장항목별 지급 요건 및 필수 제출 서류 규칙과 교차 분석함으로써, 현재 청구 중인 보장항목의 예상 보험금을 산출하는 동시에 아직 청구되지 않은 보장항목 중 보험금 지급 가능성이 있는 항목을 자동으로 발굴하는 기술에 관한 것이다. 본 발명에 따른 시스템은 약관으로부터 보장항목 목록, 가입금액, 감액 기간, 보장항목별 필수 제출 서류 목록 및 지급 요건을 추출·구조화하는 계약정보 처리부, 청구 서류로부터 의료코드, 수술 정보, 입원 기간 및 업로드된 서류 목록을 추출하는 청구서류 처리부, 규칙 일치 정도, 서류 부족 정도, OCR 인식 신뢰도 및 계약일과 진단일 또는 수술일 사이의 경과기간 등을 고려하는 예측 모델을 이용하여 각 보장항목의 보험금 지급 확률과 예상 추가 보험금을 계산하고 부족 서류를 포함하는 추천 서류 목록과 그 필요 사유를 도출하는 미지급 항목 발굴 및 서류 추천부, 그리고 기본 예상 보험금, 추가 청구 대상 보장항목, 각 항목별 예상 추가 보험금, 전체 추가 보험금 합계 및 추천 서류 목록을 리포트 형태로 사용자 단말에 제공하는 결과 제공부를 포함한다. 이에 따라 사용자는 인지하지 못한 보험금 청구 가능 항목과 부족 서류를 직관적으로 확인할 수 있고, 보험사는 자동화된 선별·검증 정보를 활용하여 심사 효율을 향상시킬 수 있어, 보험금 청구 및 심사 프로세스 전반의 편의성과 신뢰성을 높일 수 있다. The present invention relates to a system and method for identifying eligible insurance claims and recommending documents. More specifically, the present invention relates to a technology that utilizes optical character recognition (OCR) and natural language processing (NLP) to structure claim documents, such as diagnosis certificates, detailed medical expense statements, surgical certificates, and hospitalization/discharge certificates uploaded from user terminals, and insurance contract information and full terms and conditions stored on one or more insurance company servers. This technology then cross-analyzes these documents with payment requirements and required document submission rules for each covered item extracted from the terms and conditions, thereby calculating expected insurance benefits for currently claimed claims and automatically identifying potentially payable claims among unclaimed claims. The system according to the present invention includes a contract information processing unit that extracts and structures a list of covered items, subscription amount, reduction period, a list of required documents for each covered item, and payment requirements from the terms and conditions; a claim document processing unit that extracts medical codes, surgery information, hospitalization period, and a list of uploaded documents from the claim documents; a predictive model that considers the degree of rule conformance, degree of document insufficiency, OCR recognition reliability, and the elapsed time between the contract date and the date of diagnosis or surgery, and a non-payment item identification and document recommendation unit that calculates the probability of insurance payment and expected additional insurance payment for each covered item and derives a list of recommended documents including insufficient documents and the reasons for their necessity; and a result provision unit that provides a report to the user terminal containing the basic expected insurance payment, covered items subject to additional claims, the expected additional insurance payment for each item, the total additional insurance payment sum, and a list of recommended documents. Accordingly, users can intuitively identify unrecognized claimable items and insufficient documents, and insurance companies can improve screening efficiency by utilizing automated screening and verification information, thereby enhancing the convenience and reliability of the overall insurance claim and screening process.","assignee":"조상수","inventors":["조상수"],"publication_date":"2025-12-09","filing_date":"2025-11-22","priority_date":"2025-11-22","cpc_codes":["G","G06","G06Q","G06Q40/00","G06Q40/08","G06Q40/084","G06Q40/0841","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06Q","G06Q40/00","G06Q40/08","G06Q40/09"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250172913A/en"},{"publication_number":"KR20250172774A","title":"Edge AI-Based Real-Time Event Analysis System and Method for Vehicles","abstract":"본 발명은 차량용 엣지 인공지능 기반 실시간 이벤트 분석 시스템 및 방법에 관한 것으로, 차량으로부터 운행 데이터를 수집하는 데이터 수집부와, 수집된 운행 데이터를 인공지능 모델을 이용하여 분석하여 운전 이벤트를 검출하는 분석부를 포함하며, 상기 분석부는 엣지 디바이스에서 실행된다. 본 발명에 따르면, 엣지 디바이스에서 실시간으로 인공지능 모델을 실행함으로써 즉각적인 운전 이벤트 검출이 가능하고, 네트워크 트래픽을 70-90% 감소시켜 통신 비용을 절감할 수 있으며, 경량화된 인공지능 모델을 사용하여 제한된 컴퓨팅 자원에서도 95% 이상의 높은 정확도를 달성할 수 있다. The present invention relates to a real-time event analysis system and method based on edge artificial intelligence for vehicles, comprising: a data collection unit for collecting driving data from a vehicle; and an analysis unit for analyzing the collected driving data using an artificial intelligence model to detect driving events, wherein the analysis unit is executed on an edge device. According to the present invention, by executing the artificial intelligence model in real time on the edge device, immediate driving event detection is possible, communication costs can be reduced by reducing network traffic by 70-90%, and a high accuracy of 95% or more can be achieved even with limited computing resources by using a lightweight artificial intelligence model.","assignee":"주식회사 글렉","inventors":["강덕호","김은우"],"publication_date":"2025-12-09","filing_date":"2025-11-21","priority_date":"2025-11-21","cpc_codes":["G","G07","G07C","G07C5/00","G07C5/02","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G07","G07C","G07C5/00","G07C5/008","G","G07","G07C","G07C5/00","G07C5/08"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250172774A/en"},{"publication_number":"JOP20250304A1","title":"Adjuvants in the form of nanoemulsification for human papillomavirus vaccines","abstract":"يتعلق الاختراع الحالي بتوفير، إلى جانب أمور أخرى، تركيبة لقاح تتضمن مادة مساعدة من السكوالين في صورة مستحلب نانوي (SNE) squalene nanoemulsion والجسيمات الشبيهة بفيروس الورم الحليمي البشري HPV (VLPs) virus-like particles لنوع واحد على الأقل من فيروس الورم الحليمي البشري (HPV) human papillomavirus تم اختياره من المجموعة التي تتكون من أنواع HPV: 6، 11، 16، 18، 26، 31، 33، 35، 39، 45، 51، 52، 53، 55، 56، 58، 59، 66، 68، 73، و82. The present invention relates to providing, among other things, a vaccine formulation that includes a squalene nanoemulsion (SNE) adjuvant and HPV virus-like particles (VLPs) of at least one human papillomavirus (HPV) species selected from the group consisting of HPV species: 6, 11, 16, 18, 26, 31, 33, 35, 39, 45, 51, 52, 53, 55, 56, 58, 59, 66, 68, 73, and 82.","assignee":"Merck Sharp And Dohme Llc","inventors":["L Ahl Patrick","M Skinner Julie","Gaspar John","J Soukup Randal","Lea SULLIVAN Nicole","J Smith William"],"publication_date":"2025-12-08","filing_date":"2025-12-08","priority_date":"2023-06-09","cpc_codes":["A","A61","A61K","A61K39/00","A61K39/12","A","A61","A61K","A61K39/00","A61K39/39","A","A61","A61P","A61P31/00","A61P31/12","A61P31/20","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06V","G06V10/00","G06V10/40","G06V10/62","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/10","G06V20/13","G","G06","G06V","G06V20/00","G06V20/50","G06V20/52","G06V20/54","A","A61","A61K","A61K39/00","A61K2039/51","A61K2039/525","A61K2039/5258","A","A61","A61K","A61K39/00","A61K2039/555","A61K2039/55511","A61K2039/55566","C","C12","C12N","C12N2710/00","C12N2710/00011","C12N2710/20011","C12N2710/20023","C","C12","C12N","C12N2710/00","C12N2710/00011","C12N2710/20011","C12N2710/20031","C","C12","C12N","C12N2710/00","C12N2710/00011","C12N2710/20011","C12N2710/20034","C","C12","C12N","C12N7/00"],"country":"JO","kind":"application","source_url":"https://patents.google.com/patent/JOP20250304A1/en"},{"publication_number":"KR20250171236A","title":"Method and apparatus for inventory optimization","abstract":"본 발명은 재고 최적화 기술에 관한 것이다. 일 실시예에 따른 재고 최적화 방법은, 데이터 수집부에 의하여, 통신부를 통해 발주 서버로부터 하나 이상의 상품의 이전 주기의 발주 정보를 수집하는 동작; 상기 데이터 수집부에 의하여, 상기 통신부를 통해 복수의 공급 업체 서버로부터 상기 하나 이상의 상품에 대응하는 공급 정보를 수집하는 동작; 공급망 관리부에 의하여, 상기 공급 정보 및 상기 이전 주기의 발주 정보를 기초로 상기 하나 이상의 상품에 대응하는 공급 업체를 결정하는 동작; 데이터 처리부에 의하여, 데이터베이스에 저장된 상기 하나 이상의 상품의 발주 기록을 기초로 현재 주기의 발주 정보를 추정하는 동작; 상기 데이터 처리부에 의하여, 상기 하나 이상의 상품에 대하여 총재고관리비용을 최적화시키는 최적발주량을 계산하는 동작; 상기 데이터 처리부에 의하여, 상기 추정된 발주 정보 및 상기 최적발주량을 기초로 상기 하나 이상의 상품에 대한 현재 주기의 발주량을 결정하는 동작; 및 상기 공급망 관리부에 의하여, 상기 하나 이상의 상품에 대응하는 공급 업체 서버로 상기 하나 이상의 상품에 대한 현재 주기의 발주량을 전송하는 동작을 포함할 수 있다. The present invention relates to inventory optimization technology. In one embodiment, an inventory optimization method may include: collecting, by a data collection unit, order information for a previous cycle of one or more products from an ordering server via a communication unit; collecting, by the data collection unit, supply information corresponding to the one or more products from a plurality of supplier servers via the communication unit; determining, by a supply chain management unit, a supplier corresponding to the one or more products based on the supply information and the order information for the previous cycle; estimating, by a data processing unit, order information for a current cycle based on order records of the one or more products stored in a database; calculating, by the data processing unit, an optimal order quantity that optimizes total inventory management costs for the one or more products; determining, by the data processing unit, an order quantity for the current cycle for the one or more products based on the estimated order information and the optimal order quantity; and transmitting, by the supply chain management unit, the order quantity for the current cycle for the one or more products to a supplier server corresponding to the one or more products.","assignee":"주식회사 청담글로벌","inventors":["최석주"],"publication_date":"2025-12-08","filing_date":"2025-11-25","priority_date":"2024-03-07","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087","G06Q10/0872","G06Q10/08726","G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G","G06","G06F","G06F17/00","G06F17/10","G06F17/18","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06393","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087","G06Q10/0874","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087","G06Q10/0875","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0283"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250171236A/en"},{"publication_number":"KR20250171225A","title":"Method, computing device and for predicting patient information using tokenized ecg data","abstract":"토큰화된 심전도 데이터를 이용한 환자 정보 예측 방법, 컴퓨팅 장치 및 컴퓨터 프로그램이 제공된다. 본 개시의 다양한 실시예에 따른 토큰화된 심전도 데이터를 이용한 환자 정보 예측 방법은 컴퓨팅 장치에 의해 수행되는, 토큰화된 심전도 데이터를 이용한 환자 정보 예측 방법에 있어서, 환자의 생체 데이터를 획득하는 단계, 상기 획득된 생체 데이터를 토큰화 함에 따라 입력 데이터를 생성하는 단계 및 상기 생성된 입력 데이터를 기 학습된 딥러닝 기반의 환자 정보 예측 모델에 입력하여 상기 환자에 대한 환자 정보를 예측하는 단계를 포함한다. A method for predicting patient information using tokenized electrocardiogram data, a computing device, and a computer program are provided. A method for predicting patient information using tokenized electrocardiogram data according to various embodiments of the present disclosure is performed by a computing device, the method comprising: acquiring biometric data of a patient; generating input data by tokenizing the acquired biometric data; and inputting the generated input data into a pre-trained deep learning-based patient information prediction model to predict patient information about the patient.","assignee":"시너지에이아이 주식회사","inventors":["박준범","신태영","윤동환","김지훈"],"publication_date":"2025-12-08","filing_date":"2025-11-21","priority_date":"2024-01-15","cpc_codes":["A","A61","A61B","A61B5/00","A","A61","A61B","A61B5/00","A61B5/24","A61B5/30","A61B5/307","A61B5/308","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/318","A61B5/327","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/318","A61B5/346","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/318","A61B5/346","A61B5/349","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7275","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/50","G","G16","G16H","G16H50/00","G16H50/70"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250171225A/en"},{"publication_number":"KR20250171222A","title":"Product life position prediction system and method","abstract":"제품 수명 위치 예측 시스템이 제공된다. 본 발명의 실시예에 따른 제품 수명 위치 예측 시스템은, 특정 수명 위치를 결정하고자 하는 특정 제품에 대한 정보인 관심도 정보를 획득하는 관심도 정보 획득부; 상기 관심도 정보에 기 설정된 단위 수명 값 예측 알고리즘을 적용하여 상기 특정 제품의 예측 단위 수명 값을 획득하는 예측 단위 수명 값 획득부; 및 유사 제품군 또는 유사 지표 중 보다 높은 유사도를 가지는 요인을 비교 요인 정보로 선정하고, 상기 특정 제품의 상기 예측 단위 수명 값과 비교하여 상기 특정 제품의 특정 수명 위치를 결정하는 특정 수명 위치 결정부;를 포함한다. A product life position prediction system is provided. The product life position prediction system according to an embodiment of the present invention includes: an interest information acquisition unit that acquires interest information, which is information on a specific product for which a specific life position is to be determined; a predicted unit life value acquisition unit that applies a preset unit life value prediction algorithm to the interest information to acquire a predicted unit life value of the specific product; and a specific life position determination unit that selects a factor having a higher degree of similarity among similar product groups or similar indicators as comparison factor information and determines a specific life position of the specific product by comparing it with the predicted unit life value of the specific product.","assignee":"주식회사 스토어링크","inventors":["정용은"],"publication_date":"2025-12-08","filing_date":"2025-11-20","priority_date":"2022-08-19","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06315","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/087","G06Q10/0872","G06Q10/08726","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0202"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250171222A/en"},{"publication_number":"KR20250171212A","title":"Method, system, and non-transitory computer-readable recording medium for managing industrial sites","abstract":"본 발명의 일 태양에 따르면, 산업 현장을 관리하기 위한 방법으로서, 산업 현장에 존재하는 복수의 작업자가 소지한 디바이스가 상기 산업 현장에 설치되는 복수의 비콘(beacon)으로부터 수신한 제1 비콘 신호 정보를 획득하는 단계, 및 상기 제1 비콘 신호 정보로부터 특정되는 위치 정보를 참조하여 상기 복수의 작업자 중 보안 영역에 포함되는 작업자에게 알림 정보를 제공하는 단계를 포함하고, 상기 보안 영역은, 상기 산업 현장과 연관된 보안 요소에 대응되는 보안 영역이 학습된 보안 영역 예측 모델을 이용하여 결정되고, 상기 보안 영역은, 상기 산업 현장에서 동적으로 설정되어, 상기 산업 현장의 작업 장비가 이동하는 경우 상기 작업 장비의 이동 경로를 따라 상기 보안 영역이 함께 이동되고, 상기 제공하는 단계에서, 상기 산업 현장과 연관된 보안 요소를 기준으로 설정되는 보안 영역을 참조하여 상기 알림 정보의 제공 방식이 결정되고, 상기 보안 영역은, 계층적으로 설정되고, 상기 제공하는 단계에서, 상기 보안 영역은, 제1 보안 영역 및 상기 제1 보안 영역을 포함하면서 상기 제1 보안 영역보다 넓은 제2 보안 영역을 포함하고, 상기 복수의 작업자 중 상기 제1 보안 영역 내에 위치하고 있는 제1 작업자와 상기 복수의 작업자 중 상기 제1 보안 영역에는 위치하지 않고 상기 제2 보안 영역 내에 위치하는 제2 작업자에게 다른 알림 정보가 제공되거나 알림 정보가 제공되는 방식이 다르고, 상기 제1 보안 영역 내에 위치한 상기 제1 작업자가 소지하고 있는 적어도 일부 디바이스의 구동이 제한되고, 상기 복수의 작업자 중 상기 제1 보안 영역 및 상기 제2 보안 영역 외의 영역 내에 위치하고 있는 작업자의 안전모에서 보안 상황 발생 및 발생 지점에 관한 음성 알림이 알림 정보로서 제공되는 방법이 제공된다. According to one aspect of the present invention, a method for managing an industrial site comprises the steps of: obtaining first beacon signal information received by devices carried by a plurality of workers present at the industrial site from a plurality of beacons installed at the industrial site; and providing notification information to workers included in a security area among the plurality of workers with reference to location information specified from the first beacon signal information, wherein the security area is determined using a security area prediction model in which a security area corresponding to a security element associated with the industrial site is learned, and the security area is dynamically set at the industrial site, so that when work equipment of the industrial site moves, the security area moves together along a movement path of the work equipment; and in the providing step, a method of providing the notification information is determined with reference to a security area set based on a security element associated with the industrial site, and the security area is hierarchically set, and in the providing step, the security area includes a first security area and a second security area that includes the first security area and is wider than the first security area, and a first worker located within the first security area among the plurality of workers and the first security area among the plurality of workers are provided with the first security area. A method is provided in which different notification information is provided to a second worker located within the second security area and not within the area, or the method of providing notification information is different, the operation of at least some devices possessed by the first worker located within the first security area is restricted, and a voice notification regarding the occurrence of a security situation and the point of occurrence is provided as notification information from the safety helmet of a worker located within an area other than the first security area and the second security area among the plurality of workers.","assignee":"주식회사 더블티","inventors":["김영준","조원희"],"publication_date":"2025-12-08","filing_date":"2025-11-19","priority_date":"2024-05-08","cpc_codes":["G","G08","G08B","G08B21/00","G08B21/02","A","A42","A42B","A42B3/00","A42B3/04","A","A42","A42B","A42B3/00","A42B3/04","A42B3/30","G","G01","G01S","G01S1/00","G01S1/02","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G08","G08B","G08B31/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250171212A/en"},{"publication_number":"KR102896662B1","title":"Method for generating image based on instance layout and computing device using the same","abstract":"본 발명은 인스턴스 레이아웃 기반으로 이미지를 생성하는 방법에 관한 것으로서, 보다 상세하게는 (a) 제1 인스턴스 레이아웃 - 상기 인스턴스 레이아웃은 클래스에 따라 생성하고자 하는 오브젝트의 배치나 구조를 나타냄 - 내지 제n 인스턴스 레이아웃 - 상기 n은 1 이상의 정수임 - 을 포함하는 원본 컨트롤 이미지와 이미지 캡션이 획득되면, 컴퓨팅 장치가, 상기 원본 컨트롤 이미지 및 상기 이미지 캡션을 조건형 이미지 생성 모델에 입력하여 상기 조건형 이미지 생성 모델로 하여금 상기 이미지 캡션을 참조하여 상기 원본 컨트롤 이미지의 상기 제1 인스턴스 레이아웃 내지 상기 제n 인스턴스 레이아웃 각각에 제1 오브젝트 내지 제n 오브젝트 각각을 합성한 이니셜 합성 이미지를 생성하도록 하는 단계; 및 (b) 상기 컴퓨팅 장치가, 제1 오브젝트 디텍션 모델을 통해 상기 이니셜 합성 이미지에 대하여 오브젝트 디텍션한 결과를 참조하여 상기 제1 인스턴스 레이아웃 내지 상기 제n 인스턴스 레이아웃 중에서 오검출 또는 미검출된 인스턴스 레이아웃이 있는지를 확인하며, 오검출 및 미검출된 인스턴스 레이아웃이 없을 경우에는 상기 이니셜 합성 이미지를 완성된 합성 이미지로 생성하며, 오검출 또는 미검출된 인스턴스 레이아웃이 있을 경우에는, 상기 이니셜 합성 이미지에서의 오검출 또는 미검출된 적어도 하나의 특정 인스턴스 레이아웃들에 대응되는 적어도 하나의 서브 컨트롤 이미지들을 상기 조건형 이미지 생성 모델에 입력하여 상기 조건형 이미지 생성 모델로 하여금 상기 이미지 캡션을 참조하여 상기 이니셜 합성 이미지에서의 상기 적어도 하나의 특정 인스턴스 레이아웃들에 이에 대응되는 적어도 하나의 특정 오브젝트들을 재합성하도록 하여 상기 완성된 합성 이미지를 생성하는 단계;를 포함하는 방법이 개시된다. The present invention relates to a method for generating an image based on an instance layout, and more specifically, to a method for generating an image, comprising: (a) a step of: when an original control image and an image caption including a first instance layout, wherein the instance layout indicates the arrangement or structure of an object to be generated according to a class, and an n-th instance layout, wherein n is an integer greater than or equal to 1, are acquired, a computing device inputs the original control image and the image caption into a conditional image generation model, such that the conditional image generation model generates an initial composite image by synthesizing each of a first object to an n-th object into each of the first instance layout to the n-th instance layout of the original control image with reference to the image caption; And (b) the computing device refers to the result of object detection for the initial synthetic image through the first object detection model to check whether there is an instance layout that is misdetected or undetected among the first instance layout to the n-th instance layout, and if there is no instance layout that is misdetected or undetected, the initial synthetic image is generated as a completed synthetic image, and if there is an instance layout that is misdetected or undetected, the method includes inputting at least one sub-control image corresponding to at least one specific instance layout that is misdetected or undetected in the initial synthetic image into the conditional image generation model, and causing the conditional image generation model to re-synthesize at least one specific object corresponding to the at least one specific instance layout in the initial synthetic image with reference to the image caption, thereby generating the completed synthetic image.","assignee":"주식회사 슈퍼브에이아이","inventors":["고경렬"],"publication_date":"2025-12-08","filing_date":"2025-09-10","priority_date":"2025-09-10","cpc_codes":["G","G06","G06T","G06T11/00","G06T11/60","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06T","G06T5/00","G06T5/50","G","G06","G06T","G06T5/00","G06T5/77","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20212","G06T2207/20221","G","G06","G06T","G06T2210/00","G06T2210/12"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102896662B1/en"},{"publication_number":"KR102895516B1","title":"Electronic apparatus and method for generating blood vessel image","abstract":"본 개시의 방법은 동맥 단계에서 촬영된 제1 이미지를 획득하는 단계; 정맥 단계에서 촬영된 제2 이미지를 획득하는 단계; 제2 이미지에서 제1 이미지보다 밝아진 영역을 추출하여 특징 이미지를 생성하는 단계; 제1 이미지 및 특징 이미지에 기초하여 신장, 신동맥 및 신정맥 각각에 대한 마스크가 생성된 마스크 이미지를 생성하는 단계; 마스크 이미지에서 단절된 동맥 영역을 식별하는 단계; 및 단절된 동맥 영역을 최단거리로 연결하여 보정한 최종 혈관 이미지를 생성하는 단계를 포함할 수 있다. The method of the present disclosure may include the steps of: acquiring a first image captured at an arterial stage; acquiring a second image captured at a venous stage; generating a feature image by extracting an area brighter than the first image from the second image; generating a mask image in which masks are generated for each of a kidney, a renal artery, and a renal vein based on the first image and the feature image; identifying a disconnected arterial area in the mask image; and generating a corrected final vascular image by connecting the disconnected arterial area with the shortest distance.","assignee":"주식회사 메드에이아이","inventors":["고성제"],"publication_date":"2025-12-08","filing_date":"2025-07-31","priority_date":"2025-07-31","cpc_codes":["A","A61","A61B","A61B6/00","A61B6/50","A61B6/504","A","A61","A61B","A61B6/00","A61B6/52","A61B6/5205","A","A61","A61B","A61B6/00","A61B6/52","A61B6/5211","A61B6/5217","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06T","G06T5/00","G06T5/60","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G06","G06T","G06T7/00","G06T7/10","G06T7/174","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10072","G06T2207/10081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102895516B1/en"},{"publication_number":"LU601954B1","title":"Independent Fraud Detection System and Method for Self-Checkout Stations","abstract":"The present invention belongs to the technical field of fraud detection, discloses an independent fraud detection system and method for self-checkout stations. The system includes: a high-resolution camera system, a computer vision module, a behavior analysis module, a decision module, an alert module; the high-resolution camera system is configured to monitor a self-checkout area; the computer vision module is configured to implement a convolutional neural network; the behavior analysis module is configured to track customer actions; the decision module is configured to classify fraudulent activities; the alert module is configured to notify staff; where the independent fraud detection system operates independently of existing point-of-sale systems. The present invention employs state-of-the-art artificial intelligence technology to achieve real-time autonomous monitoring, detection. A key feature of this solution is its independent architecture, which eliminates the need for complex integration with existing point-of-sale (POS) systems of self-checkout stations, significantly reducing implementation barriers, operational disruptions.","assignee":"Bakuai As","inventors":["Usman Ahmed","Jerry Lin"],"publication_date":"2025-12-08","filing_date":"2025-06-06","priority_date":"2025-06-06","cpc_codes":["G","G06","G06Q","G06Q20/00","G06Q20/08","G06Q20/18","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q20/00","G06Q20/08","G06Q20/20","G06Q20/208","G","G06","G06Q","G06Q20/00","G06Q20/38","G06Q20/40","G06Q20/401","G06Q20/4016","G","G06","G06V","G06V20/00","G06V20/40","G06V20/44","G","G07","G07G","G07G1/00","G07G1/0036","G07G1/0045","G07G1/0054","G07G1/0063","G","G07","G07G","G07G3/00","G07G3/003"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU601954B1/en"},{"publication_number":"KR102897525B1","title":"AI-based intelligent drug detection system","abstract":"인공지능 기반 지능형 마약류 탐지 시스템이 개시된다. 이 시스템은 마약류 물질을 탐지하기 위한 라만 분광기로부터 수신된 원시 데이터를 전처리하고 특징 벡터를 추출하는 특징 추출부, 및 하나 이상의 마약류 예측 모델을 기반으로 특징 벡터를 분석하여 마약류 물질과 종류를 예측하는 마약류 예측부를 포함한다. An intelligent narcotics detection system based on artificial intelligence is disclosed. The system includes a feature extraction unit that preprocesses raw data received from a Raman spectrometer for detecting narcotics and extracts feature vectors, and a narcotics prediction unit that analyzes the feature vectors based on one or more narcotics prediction models to predict the narcotics substance and type.","assignee":"경찰대학 산학협력단","inventors":["유승진","곽성신","신강일","황광선","이영진","최옥주","신지영","이지윤","현서린","이주영","이정수","박철현","강병선"],"publication_date":"2025-12-08","filing_date":"2025-05-30","priority_date":"2025-05-30","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26","G","G01","G01N","G01N21/00","G01N21/62","G01N21/63","G01N21/65","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102897525B1/en"},{"publication_number":"KR102897352B1","title":"Intelligent Query and Response System Based on Multi-Agent Architecture for IT Infrastructure Monitoring","abstract":"본 발명은 IT 인프라 환경에서의 성능 상태, 구성 정보, 장애 원인 등을 자동으로 분석하고 자연어 응답을 생성하는 멀티 에이전트 기반 지능형 질의 응답 시스템에 관한 것이다. 사용자의 질의가 입력되면, 질의 해석 모듈이 목적과 대상 시스템을 분석하고, 이에 따라 성능 지표 조회용 메트릭 에이전트, 구성 정보 조회용 DB 에이전트, 일반 응답용 챗봇 에이전트, 원인 분석 및 대응 제공용 RAG 에이전트 중 하나 이상을 자동 호출하여 질의에 응답한다. 각 에이전트의 분석 결과는 필요 시 다른 에이전트로 자동 전달되어 협력적으로 응답 생성을 수행하며, 사용자의 질의 흐름은 컨텍스트 관리부에 저장되어 후속 질의 시 맥락 기반 처리 및 최적화에 활용된다. 이러한 구조를 통해 운영자는 복잡한 장애나 구성 문제에 대해 일일이 분석하지 않고도, 실시간으로 정확한 진단과 대응 방안을 자동 제공받을 수 있다. The present invention relates to a multi-agent based intelligent question-answering system that automatically analyzes performance status, configuration information, cause of failure, etc. in an IT infrastructure environment and generates a natural language response. When a user's query is entered, the query interpretation module analyzes the purpose and target system, and responds to the query by automatically calling one or more of the following: a metric agent for performance indicator lookup, a DB agent for configuration information lookup, a chatbot agent for general response, and a RAG agent for cause analysis and response provision. The analysis results of each agent are automatically transmitted to other agents when necessary to collaboratively generate responses, and the user's query flow is stored in the context management unit and utilized for context-based processing and optimization in subsequent queries. This structure allows operators to automatically receive accurate diagnoses and response measures in real time without having to manually analyze complex failures or configuration issues.","assignee":"(주)와치텍","inventors":["박권재"],"publication_date":"2025-12-08","filing_date":"2025-05-12","priority_date":"2025-05-12","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F11/00","G06F11/07","G06F11/0703","G06F11/079","G","G06","G06F","G06F16/00","G06F16/20","G06F16/22","G06F16/2228","G06F16/2237","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F16/00","G06F16/30","G06F16/34","G06F16/345","G","G06","G06F","G06F17/00","G06F17/10","G06F17/18","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N20/00","H","H04","H04L","H04L51/00","H04L51/02","G","G06","G06F","G06F2123/00","G06F2123/02","G","G06","G06F","G06F2201/00","G06F2201/81"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102897352B1/en"},{"publication_number":"CN121071030A","title":"Demonstration and understanding oriented large model multi-channel interaction joint output method","abstract":"The invention discloses a large model multi-channel interaction joint output method oriented to demonstration and understanding, which comprises the steps of S1, constructing a multi-channel interaction output model, enabling a large language model LLM to obtain text data through information retrieval and to divide the text data to obtain text fragment sequence data, S2 enabling a non-text mode data generation module to obtain non-text mode data through large language model LLM retrieval, enabling a multi-mode rendering module to conduct multi-channel rendering on the non-text mode data to obtain multi-channel rendering data, enabling a synchronous controller to synchronously correlate the multi-channel rendering data with the text fragment sequence data based on key position anchor point data, and S3 enabling the multi-channel interaction output model to synchronously correlate and output the multi-channel rendering data with each text fragment. The invention provides a brand new, more efficient and attractive information interaction mode, and helps users understand the content output by the large model more quickly and deeply through the cooperative display of the text and the multi-mode information.","assignee":"Zhejiang Shizizhizi Big Data Co ltd","inventors":["彭晓波","胡超","陈项世隆","李中瀚","蒋博宣","张波"],"publication_date":"2025-12-05","filing_date":"2025-11-10","priority_date":"2025-11-10","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/26","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2455","G06F16/24564","G","G06","G06F","G06F16/00","G06F16/20","G06F16/29","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06N","G06N5/00","G06N5/04","G06N5/045"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121071030A/en"},{"publication_number":"CN121070970A","title":"SQL sentence generation method and device and electronic equipment","abstract":"本发明提供了一种SQL语句的生成方法、装置和电子设备，包括：获取用户输入的信息，基于检索增强生成的知识库对用户输入的信息进行检索得到上下文信息；对用户输入的信息进行预处理，基于预处理后的信息与上下文信息生成SQL语句；基于预设的关键字黑名单、表名白名单、字段与数据库字段的映射关系对SQL语句进行验证；如果验证均通过，验证SQL语句是否正常执行；如果SQL语句正常执行，对SQL语句进行复杂计算逻辑判断、动态Python分析与沙箱化执行，确定查询结果。通过增强模型对复杂查询的理解能力与生成SQL语句的准确性，同时确保生成SQL语句的安全性和合规性，有效提升Text2SQL技术的整体性能和实用性。 This invention provides a method, apparatus, and electronic device for generating SQL statements, comprising: acquiring user input information; retrieving context information from the user input information based on a knowledge base generated by retrieval enhancement; preprocessing the user input information; generating an SQL statement based on the preprocessed information and the context information; validating the SQL statement based on a preset keyword blacklist, table name whitelist, and mapping relationships between fields and database fields; verifying whether the SQL statement executes correctly if all verifications pass; and performing complex calculation logic judgments, dynamic Python analysis, and sandboxed execution on the SQL statement to determine the query result. By enhancing the model's ability to understand complex queries and the accuracy of generated SQL statements, while ensuring the security and compliance of the generated SQL statements, this invention effectively improves the overall performance and practicality of Text2SQL technology.","assignee":"Yunjin Smart Technology Co ltd","inventors":["胡志华","张涛","蒋军君","何峰"],"publication_date":"2025-12-05","filing_date":"2025-11-10","priority_date":"2025-11-10","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2452","G06F16/24522","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/242","G06F16/243","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2455","G06F16/24553","G06F16/24558","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2455","G06F16/24564","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/248","G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6227","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121070970A/en"},{"publication_number":"CN121071761A","title":"Method, equipment and storage medium for detecting data quality of carbon footprint","abstract":"The application discloses a data quality detection method, equipment and a storage medium for carbon footprint, and belongs to the technical field of data quality detection. The method comprises the steps of obtaining data to be detected, retrieving at least one piece of domain knowledge information related to the data to be detected in a knowledge database, calling a data quality judging model, identifying data characteristic information of the data to be detected, determining data quality information of the data to be detected according to the data characteristic information, combining the data to be detected, the domain knowledge information and the data quality information to generate enhanced prompt information, inputting the enhanced prompt information into a large language model, and obtaining a quality detection result of the data to be detected, which is generated by the large language model based on the enhanced prompt information. According to the collaborative detection mechanism integrating the multi-source information and the artificial intelligence discrimination model, the accuracy and the reliability of quality control of multi-mode and heterogeneous carbon footprint data are remarkably improved.","assignee":"Shenzhen Institute of Advanced Technology of CAS","inventors":["冯威","刘杰","刘鲁静","薛登高","雷航"],"publication_date":"2025-12-05","filing_date":"2025-11-10","priority_date":"2025-11-10","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2433","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3344","G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/091","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06N","G06N5/00","G06N5/04","G06N5/045"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121071761A/en"},{"publication_number":"CN121073262A","title":"Power distribution network dynamic planning investment decision-making method and system based on deep double Q network","abstract":"本发明公开了一种基于深度双Q网络的配电网动态规划投资决策方法，属于电力系统优化技术领域。涉及配电自动化技术领域。该方法包括：获取区域配电网运行历史数据，对数据进行归一化处理，构建包含技术指标、设备状态、经济参数和环境因素的8维状态空间向量；基于深度双Q网络，创建配电网分析在线网络和目标网络，并初始化；定义离散投资动作集，基于 ‑greedy策略选择投资动作；执行动作并计算技术指标，设计融合技术指标奖励与经济成本惩罚的复合奖励函数；采用双网络结构和经验回放机制训练Q网络，实现投资策略的动态优化。本发明能够自动适应负荷预测偏差和设备老化等不确定性，在满足SAIDI、电压合格率等硬性技术约束下，显著降低全生命周期成本。 This invention discloses a dynamic planning investment decision-making method for distribution networks based on deep dual-Q networks, belonging to the field of power system optimization technology and specifically related to distribution automation technology. The method includes: acquiring historical operating data of the regional distribution network, normalizing the data, and constructing an 8-dimensional state space vector containing technical indicators, equipment status, economic parameters, and environmental factors; creating and initializing an online distribution network analysis network and a target network based on a deep dual-Q network; defining a discrete investment action set based on... The greedy strategy selects investment actions; executes these actions and calculates technical indicators; designs a composite reward function that integrates technical indicator rewards and economic cost penalties; and trains the Q-network using a dual-network structure and an experience replay mechanism to achieve dynamic optimization of the investment strategy. This invention can automatically adapt to uncertainties such as load forecast deviations and equipment aging, and significantly reduce total lifecycle costs while meeting hard technical constraints such as SAIDI and voltage qualification rate.","assignee":"Economic and Technological Research Institute of State Grid Anhui Electric Power Co Ltd","inventors":["杨欣","徐冉","王绪利","周帆","种亚林","田佳","郭汶璋","李坤"],"publication_date":"2025-12-05","filing_date":"2025-11-10","priority_date":"2025-11-10","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06313","G","G06","G06Q","G06Q50/00","G06Q50/06","Y","Y04","Y04S","Y04S10/00","Y04S10/50"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121073262A/en"},{"publication_number":"CN121073409A","title":"Laboratory detection full-flow intelligent traceability system based on reinforcement learning","abstract":"本发明公开了基于增强学习的实验室检测全流程智能溯源系统，涉及实验室检测流程智能化管理技术领域，该系统包括：状态数据采集模块；增强学习智能决策模块；溯源调度执行模块；混合架构存储与核验模块；模型动态优化模块；本发明通过引入多目标深度确定性策略梯度算法，结合系统综合资源负载率公式、动态多目标奖励函数公式及多目标动作价值函数公式，实现了资源利用与溯源精准性的双重优化，系统能够根据实时资源负载情况动态调整溯源策略，在高负载时自动切换为指纹存证模式以减少资源消耗，在低负载时则采用全量存证模式以确保溯源数据的完整性，这种智能调度机制显著提高了实验室检测全流程的效率和可靠性。 This invention discloses an intelligent traceability system for the entire laboratory testing process based on reinforcement learning, belonging to the field of intelligent management technology for laboratory testing processes. The system includes: a status data acquisition module; a reinforcement learning intelligent decision-making module; a traceability scheduling and execution module; a hybrid architecture storage and verification module; and a model dynamic optimization module. By introducing a multi-objective deep deterministic strategy gradient algorithm, combined with the system's comprehensive resource load rate formula, dynamic multi-objective reward function formula, and multi-objective action value function formula, this invention achieves dual optimization of resource utilization and traceability accuracy. The system can dynamically adjust the traceability strategy according to real-time resource load conditions. Under high load, it automatically switches to fingerprint storage mode to reduce resource consumption, while under low load, it adopts full storage mode to ensure the integrity of traceability data. This intelligent scheduling mechanism significantly improves the efficiency and reliability of the entire laboratory testing process.","assignee":"Lianyungang Customs Comprehensive Technical Center","inventors":["王恒","王静","乔柱","姜郁","秦立俊","徐师"],"publication_date":"2025-12-05","filing_date":"2025-11-10","priority_date":"2025-11-10","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/103","G","G06","G06F","G06F16/00","G06F16/20","G06F16/27","G","G06","G06F","G06F21/00","G06F21/60","G06F21/64","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5027","G06F9/505","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06312","H","H04","H04L","H04L9/00","H04L9/50"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121073409A/en"},{"publication_number":"CN121074454A","title":"Luxury true and false identification clustering method based on deep learning","abstract":"The invention discloses a luxury true and false identification clustering method based on deep learning, which relates to the technical field of computer vision and comprises the steps of minimizing feature distribution differences among different domains by utilizing an opposite domain adaptation network based on a fused fine-grained feature matrix, amplifying counterfeit difference responses at high-risk sites through a site attention mechanism to obtain an embedded feature matrix after cross-domain alignment, projecting the embedded feature matrix after cross-domain alignment onto a unit spherical manifold through L2 normalization, acquiring the spherical embedded feature matrix by combining multi-view consistency loss constraint, inputting a von mises-Fisher mixed model, and outputting an initial clustering center set and a cluster posterior probability matrix. The invention realizes the feature distribution alignment and the feature response amplification of high-risk counterfeiting details of different shooting domains, improves the consistency and the counterfeiting sensitivity of feature representation, and improves the robustness and the clustering precision of luxury product authenticity identification.","assignee":"Beijing Zhuanzhuan Spiritual Technology Co ltd","inventors":["李立涛"],"publication_date":"2025-12-05","filing_date":"2025-11-10","priority_date":"2025-11-10","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/70","G06V10/762","G06V10/763","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06Q","G06Q30/00","G06Q30/018","G06Q30/0185","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121074454A/en"},{"publication_number":"AU2025267468A1","title":"Methods and systems for automatically detecting design elements in a two-dimensional design document","abstract":"A computer-implemented method for detecting two-dimensional (2D) elements in 2D documents, the method comprising: receiving a plurality of 2D sample documents; extracting training data from the plurality of 2D sample documents, the training data including i) a plurality of 2D sample elements visually displayed on the plurality of 2D sample documents and ii) a plurality of designations stored on the plurality of 2D sample documents and associated with the plurality of 2D sample elements; and training a machine learning model with the training data by: detecting visual objects in the plurality of 2D sample documents indicative of the plurality of 2D sample elements for identifying 2D elements in 2D documents; and distinguishing visual features between the plurality of 2D sample elements displayed on the plurality of 2D sample documents for classifying each of the 2D elements identified in the 2D documents with at least one of the plurality of designations.","assignee":"Bluebeam Inc","inventors":["Bruno ALVES","Jae Min Lee"],"publication_date":"2025-12-04","filing_date":"2025-11-14","priority_date":"2019-06-06","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/10","G06F30/13","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06F","G06F16/00","G06F16/90","G06F16/93","G","G06","G06F","G06F30/00","G06F30/10","G06F30/12","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/10","G06V20/176","G","G06","G06V","G06V20/00","G06V20/70","G","G06","G06V","G06V30/00","G06V30/40","G06V30/42","G06V30/422"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025267468A1/en"},{"publication_number":"AU2025267483A1","title":"Systems and method for calculating liability of a driver of a vehicle","abstract":"MARKED-UP COPY MARKED-UP COPY Aspects of the present disclosure are related to systems, apparatus, and methods of generating or calculating liability and operational costs of a vehicle based on a driver’s handling of the vehicle are described herein. Using a combination of vehicle sensors, video input, and on-board artificial intelligence and/or machine learning algorithms, the systems and methods of the present disclosure can identify risky events performed by the driver of a vehicle and generate, calculate, and evaluate driving scores for the driver of the vehicle and send the calculations to one or more entities.","assignee":"Moter Technologies Inc","inventors":["Daniel Brooks","Michael Fischer","Kenji Fujii","Craig Lozofsky","Toshiyuki Shimamura"],"publication_date":"2025-12-04","filing_date":"2025-11-14","priority_date":"2019-05-17","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06Q","G06Q40/00","G06Q40/08","G","G06","G06Q","G06Q50/00","G06Q50/40","G","G07","G07C","G07C5/00","G07C5/008"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025267483A1/en"},{"publication_number":"AU2025267487A1","title":"Gaming activity monitoring systems and methods","abstract":"Embodiments relate to systems, methods and computer readable media for gaming monitoring. In particular, embodiments process images to determine presence of a gaming object on a gaming table in the images. Embodiments estimate postures of one or more players in the images and based on the estimated postures determine a target player associated with the gaming object among the one or more players.","assignee":"Angel Group Co Ltd","inventors":["Subhash Challa","Louis Quinn","Duc Dinh Minh Vo","Nhat Vo"],"publication_date":"2025-12-04","filing_date":"2025-11-14","priority_date":"2020-06-30","cpc_codes":["G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G06V40/161","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G","G06","G06T","G06T7/00","G06T7/0002","G","G06","G06T","G06T7/00","G06T7/10","G","G06","G06T","G06T7/00","G06T7/70","G06T7/73","G","G06","G06T","G06T7/00","G06T7/70","G06T7/73","G06T7/75","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/40","G06V20/41","G06V20/42","G","G06","G06V","G06V20/00","G06V20/40","G06V20/44","G","G06","G06V","G06V20/00","G06V20/50","G06V20/52","G","G06","G06V","G06V20/00","G06V20/60","G","G06","G06V","G06V20/00","G06V20/70","G","G06","G06V","G06V40/00","G06V40/10","G","G06","G06V","G06V40/00","G06V40/10","G06V40/103","G","G06","G06V","G06V40/00","G06V40/10","G06V40/107","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G06V40/168","G06V40/171","G","G06","G06V","G06V40/00","G06V40/20","G06V40/28","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3202","G07F17/3204","G07F17/3206","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3202","G07F17/3223","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3225","G07F17/3232","G07F17/3234","G","G07","G07F","G07F17/00","G07F17/32"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025267487A1/en"},{"publication_number":"AU2025267334A1","title":"AI system for predicting reading time and reading complexity for reviewing 2D/3D breast images","abstract":"Examples of the present disclosure describe systems and methods for predicting the reading time and/or reading complexity of a breast image. In aspects, a first set of data relating to the reading time of breast images may be collected from one or more data sources, such as image acquisition workstations, image review workstations, and healthcare professional profile data. The first set of data may be used to train a predictive model to predict/estimate an expected reading time and/or an expected reading complexity for various breast images. Subsequently, a second set of data comprising at least one breast image may be provided as input to the trained predictive model. The trained predictive model may output an estimated reading time and/or reading complexity for the breast image. The output of the trained predictive model may be used to prioritize mammographic studies or optimize the utilization of available time for radiologists.","assignee":"Hologic Inc","inventors":["Haili Chui","Adora DSOUZA","Nikolaos Gkanatsios","Ashwini Kshirsagar","Xiangwei Zhang"],"publication_date":"2025-12-04","filing_date":"2025-11-11","priority_date":"2019-09-27","cpc_codes":["G","G16","G16H","G16H40/00","G16H40/20","A","A61","A61B","A61B6/00","A61B6/50","A61B6/502","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06311","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06398","G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/109","G06Q10/1097","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G","G16","G16H","G16H10/00","G16H10/20","G","G16","G16H","G16H30/00","G16H30/20","G","G16","G16H","G16H30/00","G16H30/40","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/70","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","A","A61","A61B","A61B6/00","A61B6/46","A61B6/461","A61B6/465","G","G06","G06T","G06T2200/00","G06T2200/24","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30004","G06T2207/30068"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025267334A1/en"},{"publication_number":"EP4657695A1","title":"Learning a surrogate model of an hpp energy management system and use for hpp sizing","abstract":"The disclosure concerns a method, for learning an EMS surrogate. The method comprises obtaining a dataset of training examples and training the surrogate. Each example includes an input and an output. The input includes ratios representing a HPP configuration, including a ratio between a renewable production capacity and a grid connection. The ratios further include a ratio between a rated battery power and the grid connection. The ratios further include a ratio between a battery energy capacity and the rated battery power. The input further includes data representing a plurality of time series each representing renewable energy resource power over a predetermined period and a plurality of time series each representing grid energy demand over the predetermined period. The output includes data representing one or more pluralities of time series each representing a respective operational parameter of the HPP, and each obtained by applying a high-fidelity EMS to the input.","assignee":"Totalenergies Onetech; TotalEnergies Onetech SAS","inventors":["Charbel Assaad","Juan Pablo Murcia Leon","Kaushik Das","Poul Ejnar Sørensen","Sami Ghazouani"],"publication_date":"2025-12-03","filing_date":"2024-06-02","priority_date":"2024-06-02","cpc_codes":["H","H02","H02J","H02J3/00","H02J3/28","H02J3/32","G","G06","G06N","G06N3/00","G06N3/02","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","H","H02","H02J","H02J3/00","H02J3/38","H02J3/381","H","H02","H02J","H02J7/00","H02J7/34"],"country":"EP","kind":"application","source_url":"https://patents.google.com/patent/EP4657695A1/en"},{"publication_number":"CN121050713A","title":"Methods, apparatus, media, electronic equipment and products for determining code adoption rate","abstract":"本申请公开了一种代码采纳率的确定方法和装置、介质、电子设备及产品，涉及软件工程领域，包括：获取分布式版本控制系统在第一时间范围内的第一变更代码，并获取智能编程助手生成的与所述第一变更代码对应的第一响应代码；在所述第一响应代码中，确定与所述第一变更代码中的每行代码相关的候选集，并通过所述第一变更代码中的每行代码和所述候选集确定所述第一变更代码和所述第一响应代码之间具有同源关系的代码对；通过所述代码对确定对所述第一响应代码的代码采纳率。通过本申请，解决了对人工智能在软件开发过程中所生成代码的实际贡献难以准确量化评估的问题，实现了对人工智能生成的代码的贡献准确评估的效果。 This application discloses a method, apparatus, medium, electronic device, and product for determining code adoption rate, relating to the field of software engineering. The method includes: acquiring first modified code from a distributed version control system within a first time frame, and acquiring first response code generated by an intelligent programming assistant corresponding to the first modified code; determining a candidate set related to each line of code in the first modified code within the first response code, and determining code pairs with a common origin relationship between the first modified code and the first response code through each line of code in the first modified code and the candidate set; and determining the code adoption rate for the first response code through the code pairs. This application solves the problem of accurately quantifying and evaluating the actual contribution of code generated by artificial intelligence in software development, achieving an accurate evaluation of the contribution of AI-generated code.","assignee":"Inspur Jinan data Technology Co ltd; Zhengzhou Inspur Data Technology Co Ltd","inventors":["亓开元","郭立民","郭涛","刘元松","孔维亭","张霄炜"],"publication_date":"2025-12-02","filing_date":"2025-11-05","priority_date":"2025-11-05","cpc_codes":["G","G06","G06F","G06F8/00","G06F8/30","G06F8/36","G","G06","G06F","G06F8/00","G06F8/30","G06F8/34","G","G06","G06F","G06F8/00","G06F8/40","G06F8/41","G06F8/44","G","G06","G06F","G06F8/00","G06F8/70","G06F8/71","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121050713A/en"},{"publication_number":"CN121052466A","title":"A method and system for real-time data monitoring of energy utilization in a printing press workshop","abstract":"The invention provides a method and a system for monitoring energy utilization of real-time data in a printing machine workshop, which relate to the technical field of energy monitoring of printing equipment and comprise the steps of establishing a multi-layer heterogeneous sensing network to collect operation parameters, optimizing sensor deployment by adopting a particle swarm algorithm, carrying out data fusion by utilizing Kalman filtering, establishing an energy utilization evaluation model, carrying out multi-objective optimization calculation based on a genetic algorithm, and constructing an energy transmission network by combining a graph theory algorithm to realize collaborative optimization. The invention can obviously improve the energy utilization efficiency of the printing equipment, reduce the energy consumption and simultaneously ensure the production efficiency and the product quality.","assignee":"Zhejiang Meige Machinery Co ltd","inventors":["刘颖","刘国方","吴钦伟","孙宪贵","孙立俊"],"publication_date":"2025-12-02","filing_date":"2025-11-05","priority_date":"2025-11-05","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06Q","G06Q50/00","G06Q50/04","G","G06","G06Q","G06Q50/00","G06Q50/06"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121052466A/en"},{"publication_number":"CN121051880A","title":"A Reliability Analysis Method for Multi-Electric Aircraft Systems Based on FG-DBN","abstract":"The embodiment of the invention discloses a reliability analysis method for a multi-electric aircraft airborne system based on FG-DBN, which relates to the field of reliability design and can dynamically analyze the reliability of a complex polymorphic system in multi-electric aircraft airborne equipment under the condition of fuzzy uncertainty. The method comprises the steps of establishing a DBN model aiming at a complex multi-state system in airborne equipment, establishing a fuzzy gradient function model containing a fuzzy supporting radius, wherein nodes in the DBN model correspond to fault states of components in the complex multi-state system, the fuzzy gradient function model comprises a fuzzy membership model of the nodes in corresponding time slices and a state transition matrix corresponding to the fault modes, establishing a conditional probability table aiming at fault relations among the nodes in the DBN model, and identifying weak links in the complex multi-state system and analyzing reliability by using output results of the fuzzy gradient function model containing the fuzzy supporting radius and the established conditional probability table.","assignee":"Nanjing University of Aeronautics and Astronautics","inventors":["陈嘉宇","王旭航","陆钦华","马煜程","葛红娟","钱小燕"],"publication_date":"2025-12-02","filing_date":"2025-11-05","priority_date":"2025-11-05","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/10","G06F30/15","G","G06","G06F","G06F30/00","G06F30/20","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06F","G06F2111/00","G06F2111/08","G","G06","G06F","G06F2119/00","G06F2119/02","G","G06","G06F","G06F2119/00","G06F2119/12"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121051880A/en"},{"publication_number":"KR20250167547A","title":"Apparatus for predicting integrated loading data of asset management system","abstract":"본 발명은 자산관리 시스템의 통합 부하데이터 예측 장치는, 전력설비에 대한 과거 확보된 부하데이터와 부하관련 데이터를 입력받는 데이터 입력부; 데이터 입력부로부터 입력된 부하데이터를 기반으로 미확보 부하데이터를 예측하기 위한 딥러닝 모델링 및 학습조건에 따라 시뮬레이션을 실행하여 미확보 부하데이터를 예측하는 부하데이터 예측부; 및 부하데이터 예측부에서 예측된 부하데이터를 출력하는 예측결과 출력부;를 포함하는 것을 특징으로 한다. The present invention is characterized in that the integrated load data prediction device of the asset management system includes a data input unit that receives past secured load data and load-related data for power facilities; a load data prediction unit that predicts unsecured load data by executing a simulation according to deep learning modeling and learning conditions for predicting unsecured load data based on the load data input from the data input unit; and a prediction result output unit that outputs the load data predicted by the load data prediction unit.","assignee":"한국전력공사","inventors":["황재상","권규범","문성덕","박석지"],"publication_date":"2025-12-01","filing_date":"2025-11-17","priority_date":"2022-06-08","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/06","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0259","G05B23/0283","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","Y","Y04","Y04S","Y04S40/00","Y04S40/20"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250167547A/en"},{"publication_number":"KR20250167537A","title":"Apparatus for evaluating health index of power distribution equipment","abstract":"본 발명은 배전 자산의 생애 주기 데이터를 포함하는 배전 자산 데이터를 수집하는 데이터 수집부, 데이터 수집부를 통해 수집된 배전 자산 데이터에 기초하여 배전 자산의 잔존 수명을 예측하는 잔존 수명 예측부, 데이터 수집부를 통해 수집된 배전 자산 데이터와 잔존 수명 예측부를 통해 예측된 배전 자산의 잔존 수명에 기초하여 배전 자산의 잔존 수명에 영향을 미치는 유효 인자를 추출하는 유효 인자 추출부, 및 잔존 수명 예측부를 통해 예측된 배전 자산의 잔존 수명과 유효 인자 추출부를 통해 추출된 유효 인자에 기초하여 배전 자산의 건전도를 나타내는 건전도 지수를 산출하는 건전도 지수 산출부를 포함하는 것을 특징으로 한다. The present invention is characterized by including a data collection unit that collects distribution asset data including life cycle data of distribution assets, a remaining life prediction unit that predicts the remaining life of the distribution asset based on the distribution asset data collected through the data collection unit, an effective factor extraction unit that extracts an effective factor affecting the remaining life of the distribution asset based on the distribution asset data collected through the data collection unit and the remaining life of the distribution asset predicted through the remaining life prediction unit, and a health index calculation unit that calculates a health index indicating the health of the distribution asset based on the remaining life of the distribution asset predicted through the remaining life prediction unit and the effective factor extracted through the effective factor extraction unit.","assignee":"한국전력공사","inventors":["이병성","이혜선","이승호","박석지","조현창"],"publication_date":"2025-12-01","filing_date":"2025-11-13","priority_date":"2020-09-22","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/06","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0218","G05B23/0224","G05B23/024","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0259","G05B23/0283","G","G06","G06F","G06F17/00","G06F17/10","G06F17/18","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0278"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250167537A/en"},{"publication_number":"KR20250167535A","title":"Artificial neural network training using flexible floating point tensors","abstract":"그러므로, 본 개시내용은 텐서를 이용하여 신경 네트워크들을 트레이닝하는 시스템 및 방법에 관한 것이고, 이 텐서는 복수의 FP16 값 및 텐서에 포함된 FP16 값들 중 일부 또는 전부에 의해 공유되는 지수를 정의하는 복수의 비트를 포함한다. FP16 값들은 IEEE 754 포맷 16-비트 부동 소수점 값들을 포함할 수 있고, 텐서는 공유된 지수를 정의하는 복수의 비트를 포함할 수 있다. 텐서는 공유된 지수 및 FP16 값들을 포함할 수 있고, FP16 값들은 프로세서 회로에 의해 동적으로 설정될 수 있는 가변 비트-길이 가수 및 가변 비트-길이 지수를 포함한다. 텐서는 공유된 지수 및 FP16 값들을 포함할 수 있고, FP16 값들은 프로세서 회로에 의해 동적으로 설정될 수 있는 가변 비트-길이 가수; 가변 비트-길이 지수; 및 FP16 값 지수를 공유된 지수와 선택적으로 조합하도록 프로세서 회로에 의해 설정되는 공유된 지수 스위치를 포함한다. Therefore, the present disclosure relates to systems and methods for training neural networks using a tensor, the tensor comprising a plurality of FP16 values and a plurality of bits defining an exponent shared by some or all of the FP16 values included in the tensor. The FP16 values may comprise IEEE 754 format 16-bit floating point values, and the tensor may comprise a plurality of bits defining the shared exponent. The tensor may comprise the shared exponent and the FP16 values, wherein the FP16 values comprise a variable bit-length mantissa and a variable bit-length exponent that can be dynamically set by a processor circuit. The tensor may comprise the shared exponent and the FP16 values, wherein the FP16 values comprise a variable bit-length mantissa that can be dynamically set by the processor circuit; a variable bit-length exponent; and a shared exponent switch configured by the processor circuit to selectively combine the FP16 value exponent with the shared exponent.","assignee":"인텔 코포레이션","inventors":["크리쉬나쿠마르 네어","앤드류 양","브라이언 모리스"],"publication_date":"2025-12-01","filing_date":"2025-11-13","priority_date":"2018-06-08","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/499","G06F7/49905","G06F7/4991","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/499","G06F7/49942","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/30003","G06F9/30007","G06F9/30025","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/30098","G06F9/3012","G06F9/3013","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06F","G06F2207/00","G06F2207/38","G06F2207/3804","G06F2207/3808","G06F2207/3812","G06F2207/3816","G","G06","G06F","G06F2207/00","G06F2207/38","G06F2207/3804","G06F2207/3808","G06F2207/3812","G06F2207/382"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250167535A/en"},{"publication_number":"MX2025013560A","title":"Display method and electronic device","abstract":"Provided in the present application are a display method and an electronic device. The method can be applied to an electronic device, wherein a first display interface of the electronic device comprises an entry identifier used for calling an AI service. The method comprises: performing detection on a first operation for an entry identifier; and in response to the first operation, displaying a second display interface used for interacting with an AI service. In the technical solution, a display interface of an electronic device may comprise an entry identifier used for calling an AI service, and a user may call, by means of operating the entry identifier, a display interface used for interacting with the AI service, and therefore the user can conveniently use the AI service in the electronic device.","assignee":"Huawei Tech Co Ltd","inventors":["Hongjun Wang","Yifan Chen","Yanan Zhang","Wooseok Hwang","Yankun Tai","Yuxiao Zhou"],"publication_date":"2025-12-01","filing_date":"2025-11-12","priority_date":"2023-07-18","cpc_codes":["G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/451","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481","G06F3/0482","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G06F3/04845","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G06F3/04847","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0487","G06F3/0488","G06F3/04883","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0487","G06F3/0488","G06F3/04886","G","G06","G06F","G06F3/00","G06F3/16","G06F3/167","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06F","G06F2203/00","G06F2203/048","G06F2203/04803"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2025013560A/en"},{"publication_number":"MX2025013434A","title":"Industrial system","abstract":"The invention relates to determining actions to improve operation of an industrial system. Reference data indicative of one or more states of the industrial system is obtained and it is determined whether a deviation is present within the reference data. If a deviation is present, the effect of the deviation on the industrial system is assessed and one or more mitigating control actions are selection from a plurality of predetermined control actions based on the assessed effect of the deviation. A representation of the impact of performing one or more mitigating control actions is output to a user interface, enabling a user of the user interface to select an operation to initiate one or more mitigating control actions to change one or more processes of the industrial system.","assignee":"Gea Group Ag","inventors":["Hassan Yazdi","Kevin Feldmann","Lukas Roy Svane Theisen"],"publication_date":"2025-12-01","filing_date":"2025-11-10","priority_date":"2023-05-11","cpc_codes":["G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0218","G05B23/0243","G05B23/0254","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0208","G05B23/0216","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0218","G05B23/0224","G05B23/024","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0259","G05B23/0286","G05B23/0294","G","G06","G06N","G06N20/00"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2025013434A/en"},{"publication_number":"MX2025013304A","title":"Identification and correction of weld paths using optical coherence tomography","abstract":"A method and system tor laser welding a workpiece. The system may include a laser source emitting processing laser radiation, an OCT system that includes an imaging light source emitting imaging light and is configured to generate an interferometric output, a laser head configured to direct processing laser radiation and a beam of the imaging light onto the workpiece, at least one scanner configured to direct the beam of imaging light along at least one measurement weld path, on the workpiece, and at least one controIler configured to: receive data corresponding to a reference weld path on the workpiece, generate a corrected weld path, based at least in part on a comparison between the interferometric output of the at least one measurement weld path and the reference weld path, and control the laser head such that the processing laser radiation is directed along the corrected weld path on the workpiece.","assignee":"Ipg Photonics Corp","inventors":["Nicholas D Schwenger","Kodie D Becker","Paul J L Webster"],"publication_date":"2025-12-01","filing_date":"2025-11-06","priority_date":"2023-05-17","cpc_codes":["B","B23","B23K","B23K26/00","B23K26/02","B23K26/03","B","B23","B23K","B23K26/00","B23K26/02","B23K26/03","B23K26/032","B","B23","B23K","B23K26/00","B23K26/02","B23K26/04","B23K26/044","B","B23","B23K","B23K26/00","B23K26/08","B23K26/082","B","B23","B23K","B23K26/00","B23K26/20","B23K26/21","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2025013304A/en"},{"publication_number":"KR20250166823A","title":"Method and devices for 3-dimensional fabric draping simulation","abstract":"3차원 천의 착장 시뮬레이션 방법은 복수의 물성(physical property) 파라미터들을 입력하는 사용자 인터페이스(user interface)를 통한 사용자 입력에 기초하여 복수의 물성 파라미터들을 조정하고, 복수의 물성 파라미터들을 신경망에 인가함으로써 생성된 메쉬에 기초하여 물성 파라미터들에 대응하는 천이 미리 정해진 오브젝트에 착장된 3차원 형상을 출력하는 단계들을 포함한다. A three-dimensional cloth wearing simulation method includes steps of adjusting a plurality of physical property parameters based on user input through a user interface for inputting a plurality of physical property parameters, and outputting a three-dimensional shape in which cloth corresponding to the physical property parameters is worn on a predetermined object based on a mesh generated by applying the plurality of physical property parameters to a neural network.","assignee":"(주)클로버추얼패션","inventors":["주은정","최명걸","심응준"],"publication_date":"2025-11-28","filing_date":"2025-11-21","priority_date":"2021-12-17","cpc_codes":["G","G06","G06T","G06T19/00","G06T19/003","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G06F3/04845","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G06F3/04847","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06T","G06T15/00","G06T15/04","G","G06","G06T","G06T17/00","G06T17/20","G","G06","G06T","G06T19/00","G06T19/20","G","G06","G06T","G06T7/00","G06T7/10","G06T7/13","G","G06","G06T","G06T7/00","G06T7/97","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481","G06F3/04815","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G06F3/04842","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/067","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/067","G06N3/0675","G","G06","G06T","G06T2200/00","G06T2200/24","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2210/00","G06T2210/16","G","G06","G06T","G06T2210/00","G06T2210/36","G","G06","G06T"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250166823A/en"},{"publication_number":"KR20250166806A","title":"An artificial intelligence apparatus for detecting defective products based on product images and method thereof","abstract":"본 개시의 실시 예에 따른 인 공 지능 장치는 정상 분류에 속하는 적어도 하나의 정상 제품 이미지 및 비정상 분류에 속하는 적어도 하나의 비정상 제품 이미지를 저장하는 메모리, 동일 분류에 속하는 제품 이미지의 표현 벡터(Representation Vector)가 가까워지도록 하고, 서로 다른 분류에 속하는 제품 이미지의의 표현 벡터가 서로 멀어지도록 특징 추출 모델에 대한 대조 학습(Contrastive Leaning)을 시키는 러닝 프로세서 및 상기 대조 학습된 특징 추출 모델에 상기 정상 분류에 속하는 적어도 하나의 정상 제품 이미지를 입력하여 적어도 하나의 정상 제품 이미지의 패치 단위별 임베딩 벡터(embedding vector)를 획득하고, 상기 획득한 패치 단위별 임베딩 벡터의 정규 분포를 획득하는 프로세서를 포함한다. An artificial intelligence device according to an embodiment of the present disclosure includes a memory that stores at least one normal product image belonging to a normal classification and at least one abnormal product image belonging to an abnormal classification, a learning processor that performs contrastive learning on a feature extraction model so that representation vectors of product images belonging to the same classification become closer to each other and representation vectors of product images belonging to different classifications become farther from each other, and a processor that inputs at least one normal product image belonging to the normal classification into the contrastively learned feature extraction model to obtain an embedding vector for each patch of at least one normal product image, and obtains a normal distribution of the obtained embedding vector for each patch.","assignee":"주식회사 Lg 경영개발원","inventors":["김상윤","강병준","고영산","현지호","김승환"],"publication_date":"2025-11-28","filing_date":"2025-11-14","priority_date":"2022-03-23","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T7/00","G","G06","G06V","G06V10/00","G06V10/40","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30108","G06T2207/30164"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250166806A/en"},{"publication_number":"KR20250166807A","title":"An artificial intelligence apparatus for detecting defective products based on product images and method thereof","abstract":"본 개시의 일 실시 예에 따른 인공 지능 장치는 정상 제품 이미지를 저장하는 메모리, 정상 제품 이미지를 이미지 복원 모델에 학습 데이터로 입력하여 이미지 복원 모델이 정상 제품 이미지와 근사한 정상 복원 이미지를 출력하도록 이미지 복원 모델을 학습시키는 러닝 프로세서, 및 정상 제품 이미지를 변형하여 정상 분류에 속하는 정상 변형 이미지를 생성하고 정상 분류에 속하는 정상 제품 이미지를 증가시키고, 정상 분류에 속하는 정상 제품 이미지를 변형하여 비정상 분류에 속하는 비정상 변형 이미지를 생성하고, 비정상 변형 이미지를 이미지 복원 모델로 입력하여 이미지 복원 모델로부터 출력되는 비정상 복원 이미지를 획득하는 프로세서를 포함하고, 러닝 프로세서는 소정의 이미지 데이터를 입력받고 입력된 소정의 이미지 데이터에 대한 표현 벡터를 출력하는 특징 추출 모델에 대하여, 정상 분류에 속하는 정상 제품 이미지, 비정상 변형 이미지 및 비정상 복원 이미지를 특징 추출 모델에 입력하여, 특징 추출 모델로부터 출력되는 정상 분류에 속하는 정상 제품 이미지의 표현 벡터 및 비정상 복원 이미지의 표현 벡터간의 거리가 가까워지고, 비정상 변형 이미지의 표현 벡터 및 비정상 복원 이미지의 표현 벡터간의 거리가 멀어지도록 특징 추출 모델을 대조 학습(Contrastive Learning)시킬 수 있다. An artificial intelligence device according to one embodiment of the present disclosure includes a memory that stores a normal product image, a learning processor that inputs the normal product image as learning data into an image restoration model and trains the image restoration model to output a normal restored image similar to the normal product image, and a processor that transforms the normal product image to generate a normal transformed image belonging to a normal category and increases the number of normal product images belonging to the normal category, transforms the normal product image belonging to the normal category to generate an abnormal transformed image belonging to an abnormal category, and inputs the abnormal transformed image into the image restoration model to obtain an abnormal restored image output from the image restoration model, wherein the learning processor inputs a normal product image belonging to the normal category, an abnormal transformed image, and an abnormal restored image into a feature extraction model that receives predetermined image data and outputs an expression vector for the input predetermined image data, and performs contrastive learning on the feature extraction model such that a distance between an expression vector of a normal product image belonging to the normal category and an expression vector of an abnormal restored image output from the feature extraction model becomes closer, and a distance between an expression vector of an abnormal transformed image and an expression vector of an abnormal restored image becomes farther.","assignee":"주식회사 Lg 경영개발원","inventors":["강병준","김상윤","고영산","현지호","김승환"],"publication_date":"2025-11-28","filing_date":"2025-11-14","priority_date":"2022-03-23","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06T","G06T5/00","G06T5/60","G","G06","G06V","G06V10/00","G06V10/40","G","G06","G06V","G06V10/00","G06V10/40","G06V10/46","G06V10/469","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30108"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250166807A/en"},{"publication_number":"CO2025015811A2","title":"Decoder, encoder, decoding method, and encoding method","abstract":"Un decodificador (200) incluye memoria (252) y circuito (251) acoplado a la memoria (252). Usando la memoria (252), el circuito (251): decodifica, a partir de un flujo de bits, una unidad de datos de base de una imagen facial relacionada con un video facial y una o más unidades de datos de mejora de la imagen facial; decodifica, a partir del flujo de bits, información geométrica correspondiente a cada uno de los cuadros del video facial; y genera el video facial a partir de la unidad de datos de base, la una o más unidades de datos de mejora y la información geométrica. En el flujo de bits, la unidad de datos de base se agrega a un conjunto de datos que corresponde a un primer cuadro que es un cuadro del video facial. En el flujo de bits, la una o más unidades de datos de mejora se agregan a uno o más conjuntos de datos que corresponden a uno o más segundos cuadros del video facial. A decoder (200) includes memory (252) and circuitry (251) coupled to the memory (252). Using the memory (252), circuitry (251): decodes, from a bitstream, a base data unit of a facial image related to a facial video and one or more enhancement data units of the facial image; decodes, from the bitstream, geometric information corresponding to each frame of the facial video; and generates the facial video from the base data unit, the one or more enhancement data units, and the geometric information. In the bitstream, the base data unit is added to a data set corresponding to a first frame of the facial video. In the bitstream, the one or more enhancement data units are added to one or more data sets corresponding to one or more second frames of the facial video.","assignee":"Panasonic Ip Corp America","inventors":["Han Boon Teo","Chong Soon Lim","Sugiri Pranata Lim","Jayashree KARLEKAR","Jing Yuan Thong","Kiyofumi Abe","Takahiro Nishi","Tadamasa Toma"],"publication_date":"2025-11-28","filing_date":"2025-11-14","priority_date":"2023-05-25","cpc_codes":["H","H04","H04N","H04N19/00","H04N19/10","H04N19/134","H04N19/167","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/103","H04N19/105","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/117","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/119","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/124","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/132","H","H04","H04N","H04N19/00","H04N19/10","H04N19/134","H04N19/136","H","H04","H04N","H04N19/00","H04N19/10","H04N19/134","H04N19/157","H04N19/159","H","H04","H04N","H04N19/00","H04N19/10","H04N19/134","H04N19/162","H","H04","H04N","H04N19/00","H04N19/10","H04N19/169","H04N19/17","H","H04","H04N","H04N19/00","H04N19/10","H04N19/169","H04N19/17","H04N19/172","H","H04","H04N","H04N19/00","H04N19/20","H","H04","H04N","H04N19/00","H04N19/20","H04N19/23","H","H04","H04N","H04N19/00","H04N19/30","H","H04","H04N","H04N19/00","H04N19/30","H04N19/33","H","H04"],"country":"CO","kind":"application","source_url":"https://patents.google.com/patent/CO2025015811A2/en"},{"publication_number":"KR20250166804A","title":"Method for predicting malfunction of optical module and apparuatus thereof","abstract":"본 발명은 컴퓨팅 장치에 의해 수행되는 광모듈 장애 예측 방법에 관한 것으로, 광통신망을 구성하는 광모듈과 관련된 광레벨 데이터를 획득하는 단계; 상기 획득된 광레벨 데이터를 전처리하는 단계; 시계열 분석 알고리즘을 기반으로 상기 전처리된 광레벨 데이터를 학습하여 광레벨 추이 예측 모델을 생성하는 단계; 상기 광레벨 추이 예측 모델을 이용하여 미래의 광레벨 데이터를 예측하는 단계; 및 상기 예측된 광레벨 데이터와 미리 설정된 기준값을 비교하여 상기 광모듈의 장애 발생 여부를 예측하는 단계를 포함한다. The present invention relates to a method for predicting an optical module failure performed by a computing device, comprising: a step of obtaining optical level data related to an optical module constituting an optical communication network; a step of preprocessing the obtained optical level data; a step of learning the preprocessed optical level data based on a time series analysis algorithm to create a light level trend prediction model; a step of predicting future optical level data using the light level trend prediction model; and a step of predicting whether a failure of the optical module has occurred by comparing the predicted optical level data with a preset reference value.","assignee":"주식회사 케이티","inventors":["김영준","김은도","예우석","이진하","조광현"],"publication_date":"2025-11-28","filing_date":"2025-11-13","priority_date":"2023-12-18","cpc_codes":["H","H04","H04B","H04B10/00","H04B10/07","H04B10/075","H04B10/077","H04B10/0773","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","H","H04","H04L","H04L41/00","H04L41/06","H04L41/0631","H","H04","H04L","H04L41/00","H04L41/14","H04L41/145","H","H04","H04L","H04L41/00","H04L41/14","H04L41/147","H","H04","H04L","H04L41/00","H04L41/16"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250166804A/en"},{"publication_number":"KR20250166799A","title":"Apparatus and method of anomaly detection using neural network","abstract":"뉴럴 네트워크를 이용한 이상 검출(anomaly detection) 장치 및 방법이 개시된다. 일 실시예에 따른 이상 검출(anomaly detection) 장치는, 데이터를 수신하는 수신기, 및 정상 데이터에 대해 미리 설정된 그룹과 상기 데이터 사이의 거리에 기초하여 상기 데이터에 대응하는 복수의 특징을 추출하고, 상기 복수의 특징에 기초하여 상기 복수의 그룹에 대응하는 복수의 OOD(Out Of Distribution) 점수를 계산하고, 상기 복수의 OOD 점수에 기초하여 상기 데이터의 정상 여부를 검출하는 프로세서를 포함할 수 있다. An anomaly detection device and method using a neural network are disclosed. An anomaly detection device according to one embodiment may include a receiver that receives data, and a processor that extracts a plurality of features corresponding to the data based on a distance between a preset group for normal data and the data, calculates a plurality of OOD (Out Of Distribution) scores corresponding to the plurality of groups based on the plurality of features, and detects whether the data is normal based on the plurality of OOD scores.","assignee":"주식회사 엘로이랩","inventors":["유광선","김명환"],"publication_date":"2025-11-28","filing_date":"2025-11-13","priority_date":"2022-05-16","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250166799A/en"},{"publication_number":"KR20250166801A","title":"Apparatus for calculating health index of power distribution asset","abstract":"본 발명은 배전 자산의 생애 주기 데이터를 포함하는 배전 자산 데이터를 수집하는 데이터 수집부, 데이터 수집부를 통해 수집된 배전 자산 데이터에 기초하여 배전 자산의 잔존 수명을 예측하는 잔존 수명 예측부, 데이터 수집부를 통해 수집된 배전 자산 데이터와 잔존 수명 예측부를 통해 예측된 배전 자산의 잔존 수명에 기초하여 배전 자산의 잔존 수명에 영향을 미치는 유효 인자를 추출하는 유효 인자 추출부, 및 잔존 수명 예측부를 통해 예측된 배전 자산의 잔존 수명과 유효 인자 추출부를 통해 추출된 유효 인자에 기초하여 배전 자산의 건전도를 나타내는 건전도 지수를 산출하는 건전도 지수 산출부를 포함하는 것을 특징으로 한다. The present invention is characterized by including a data collection unit that collects distribution asset data including life cycle data of distribution assets, a remaining life prediction unit that predicts the remaining life of the distribution asset based on the distribution asset data collected through the data collection unit, an effective factor extraction unit that extracts an effective factor affecting the remaining life of the distribution asset based on the distribution asset data collected through the data collection unit and the remaining life of the distribution asset predicted through the remaining life prediction unit, and a health index calculation unit that calculates a health index indicating the health of the distribution asset based on the remaining life of the distribution asset predicted through the remaining life prediction unit and the effective factor extracted through the effective factor extraction unit.","assignee":"한국전력공사","inventors":["이병성","이혜선","이승호","박석지","조현창"],"publication_date":"2025-11-28","filing_date":"2025-11-13","priority_date":"2020-09-22","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/06","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0218","G05B23/0224","G05B23/024","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0259","G05B23/0283","G","G06","G06F","G06F17/00","G06F17/10","G06F17/18","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0278"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250166801A/en"},{"publication_number":"KR20250166794A","title":"Neural network-based artificial intelligence processor device and method of operation thereof","abstract":"본 발명은 신경망 기반 인공지능 프로세싱 장치 및 그 동작 방법에 관한 것으로, 상기 장치는 파이프라인 구조의 신경망 동작 중 적어도 인코딩 및 디코딩 연산 과정에서 접근되는 버퍼들에 있어서, 상기 인코딩 연산의 각 인코딩 시점에 순차적으로 연결된 인코딩 유닛에 의해 배타적으로 접근되는 인코더 버퍼들; 상기 디코딩 연산의 각 디코딩 시점에 상기 인코딩 유닛과 대칭되고 순차적으로 연결된 디코딩 유닛에 의해 배타적으로 접근되는 디코더 버퍼들; 및 상기 각 인코딩 시점에 상기 인코딩 유닛에 의해 출력되고 상기 각 디코딩 시점에 상기 디코딩 유닛에 의해 공유되는 연결값을 저장하는 연결 버퍼들;을 포함한다. The present invention relates to a neural network-based artificial intelligence processing device and an operating method thereof, wherein the device comprises, among buffers accessed during at least encoding and decoding operations in a neural network operation having a pipeline structure, encoder buffers exclusively accessed by encoding units sequentially connected at each encoding time point of the encoding operation; decoder buffers exclusively accessed by decoding units symmetrically and sequentially connected to the encoding units at each decoding time point of the decoding operation; and connection buffers storing connection values output by the encoding units at each encoding time point and shared by the decoding units at each decoding time point.","assignee":"세종대학교산학협력단","inventors":["박우찬","김지영","한윤호"],"publication_date":"2025-11-28","filing_date":"2025-11-12","priority_date":"2024-04-04","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06F","G06F15/00","G06F15/76","G06F15/78","G06F15/7807","G06F15/781","G","G06","G06F","G06F15/00","G06F15/76","G06F15/78","G06F15/7839","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250166794A/en"},{"publication_number":"AU2025263851A1","title":"Emergency response system","abstract":"63 Generally described, one or more aspects of the present application correspond to distributed computing systems for coordinating emergency rescue operations. The disclosed distributed emergency response system includes a mobile rescue application that increases the efficiency and reliability of users submitting rescue requests, a server-based application that coordinates dispatches in response to incoming rescue requests, and a mobile responder application that provides more efficient rescue through real-time location updates received from the mobile rescue application. The rescue application can include a dynamic, menu-based user interface to quickly solicit the emergency information needed for automated triage and dispatch.","assignee":"MosSmith Industries Inc","inventors":["Padraig MACGABANN"],"publication_date":"2025-11-27","filing_date":"2025-11-07","priority_date":"2018-04-06","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06316","G","G16","G16H","G16H40/00","G16H40/20","G","G16","G16H","G16H40/00","G16H40/60","G16H40/67","H","H04","H04L","H04L67/00","H04L67/50","H04L67/52","H","H04","H04L","H04L67/00","H04L67/50","H04L67/56","H","H04","H04L","H04L67/00","H04L67/50","H04L67/60","H04L67/61","H","H04","H04M","H04M1/00","H04M1/72","H04M1/724","H04M1/72403","H04M1/72418","H","H04","H04M","H04M3/00","H04M3/42","H04M3/50","H04M3/51","H04M3/5116","H","H04","H04W","H04W4/00","H04W4/02","H04W4/023","H","H04","H04W","H04W4/00","H04W4/02","H04W4/029","H","H04","H04W","H04W4/00","H04W4/90","H","H04","H04M","H04M2242/00","H04M2242/04","H","H04","H04M","H04M2242/00","H04M2242/30"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025263851A1/en"},{"publication_number":"AU2025263855A1","title":"System and method of utilizing data of medical systems","abstract":"The present disclosure provides a system that may receive first sensor data associated with first measurements of multiple components of respective multiple medical systems; may determine first one or more classifiers based at least on first user input; may receive second 5 sensor data associated with second measurements of the multiple components; may determine, based at least on the first one or more classifiers and based at least on the second sensor data, at least two services to be provided to respective at least a first two of the multiple medical systems; may receive third sensor data associated with third measurements of multiple components respectively associated with the at least the first two of the multiple 10 medical systems; and may determine, without the first user input and without second user input, second one or more classifiers based at least on the third sensor data.","assignee":"Alcon Inc","inventors":["Heiko Bohn","Peter Martin"],"publication_date":"2025-11-27","filing_date":"2025-11-07","priority_date":"2019-03-27","cpc_codes":["G","G06","G06K","G06K7/00","G06K7/10","G06K7/10009","G06K7/10366","G","G06","G06K","G06K7/00","G06K7/10","G06K7/14","G06K7/1404","G06K7/1408","G06K7/1413","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06311","G06Q10/063112","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06311","G06Q10/063114","G","G16","G16H","G16H40/00","G16H40/20","G","G16","G16H","G16H40/00","G16H40/40"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025263855A1/en"},{"publication_number":"US20250365139A1","title":"Quantum system for self driving cars","abstract":"A quantum method that uses the Mz quantum circuit that receives a given single chaotic value to generate strong quantum key stream of three keys. The method includes encrypting a quantum plain color image; applying set of CNOT gates between qubits of a quantum image and qubits of a quantum key to obtain a quantum encrypted image; generating a new quantum key, applying additional set of CNOT gates between the qubits the quantum color image, and the corresponding qubits of the quantum key as target qubits; performing a quantum plain scrambling (QPS) operation; and applying set of CNOT gates; and obtaining, by the quantum circuit, a quantum ciphered image.","assignee":"Abu Dhabi University","inventors":["Montasir Yousof Abdallah Qasymeh","Mohammed Abdellatif Abdelaal Zidan","Hichem El Euch","Ahmed Abdelshakour Metwaly Elfouly"],"publication_date":"2025-11-27","filing_date":"2024-05-26","priority_date":"2024-05-26","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/20","G","G06","G06N","G06N10/00","G06N10/60","H","H04","H04L","H04L9/00","H04L9/08","H04L9/0816","H04L9/0852","H","H04","H04L","H04L9/00","H04L9/08","H04L9/0861"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250365139A1/en"},{"publication_number":"US20250363377A1","title":"Compressing and transforming vector operations in an ai model","abstract":"Techniques are described herein that are capable of compressing and transforming vector operations in an AI model. First output multi-bit elements (MBEs) are generated by combining input single-bit components (SBCs) representing an input token in an AI prompt and first SBCs representing a first layer of the AI model using an exclusive-or operation. The first output MBEs are transformed into first output single-bit elements (SBEs) using a random probability distribution. Second output MBEs are generated by combining intermediate SBEs corresponding to intermediate MBEs derived from the first output SBEs and second SBCs representing a second layer of the AI model using the exclusive-or operation. A response to the AI prompt is generated to include an output token corresponding to a combination of a norm of the intermediate MBEs, a norm of second multi-bit components from which the second SBCs are derived, and a representation of the second output MBEs.","assignee":"Microsoft Technology Licensing LLC","inventors":["Omer Neter","Rayan DAHER","Daniel NISSANI"],"publication_date":"2025-11-27","filing_date":"2024-05-26","priority_date":"2024-05-26","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/091","G","G06","G06N","G06N20/00","G","G06","G06F","G06F17/00","G06F17/10","G06F17/16","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/30003","G06F9/30007","G06F9/3001","G06F9/30014","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/30003","G06F9/30007","G06F9/30018","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/30003","G06F9/30007","G06F9/30025","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/30003","G06F9/30007","G06F9/30036","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N7/00","G06N7/01"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250363377A1/en"},{"publication_number":"DE202025106945U1","title":"Federated gradient consensus framework for secure and trust-adaptive collaborative model training across distributed nodes","abstract":"Föderiertes Gradienten-Konsens-Rahmenwerk zur sicheren und vertrauensadaptiven kollaborativen Modellschulung über verteilte Knoten, umfassend: mehrere Client-Knoten, die lokales Training und Gradientenberechnung durchführen; eine föderierte Gradienten-Konsens-Schicht, die Gradientenrichtungen validiert und einen verteilten Konsens bildet; eine Vertrauensbewertungs-Einheit, die den Knoten dynamische Vertrauensgewichte zuweist; und einen sicheren Gradienten-Aggregator, der kryptografische oder Mehrparteien-Berechnungsverfahren zur datenschutzgerechten Aggregation verwendet. Federated gradient consensus framework for secure and trust-adaptive collaborative model training across distributed nodes, including: multiple client nodes that perform local training and gradient computation; a federated gradient consensus layer that validates gradient directions and forms a distributed consensus; a confidence rating unit that assigns dynamic confidence weights to the nodes; and a secure gradient aggregator that uses cryptographic or multi-party computational methods for privacy-compliant aggregation.","assignee":"Individual","inventors":[],"publication_date":"2025-11-25","filing_date":"2025-11-13","priority_date":"2025-11-13","cpc_codes":["G","G06","G06N","G06N20/00"],"country":"DE","kind":"application","source_url":"https://patents.google.com/patent/DE202025106945U1/en"},{"publication_number":"KR20250164670A","title":"AI Intelligent Photonic Lighting Platform","abstract":"본 발명은 인공지능(AI) 기반의 자율 학습 및 예측을 통해 광원의 특성과 에너지를 통합적으로 관리하고 최적화하여조명 산업의 패러다임을 근본적으로 전환하는 AI 지능형 광자 조명 플랫폼에 관한 것이다. 본 발명은 인체 및 환경에 최적화된 빛 환경 및 에너지 관리 전략을 도출하는 AI 제어 모듈, 나노광자 수준에서 빛을 정밀 조절하는 광원 모듈부, 지능형 복합 에너지 생성 및 순환 관리 시스템, 그리고 AI의 통합 관리 하에 고효율, 장수명 및 모듈형 구조를 가지며 센서와 통신 기능을 통합한 초연결 플랫폼 아키텍처를 포함한다. 본 발명은 기존 조명이 가진 발열, 비효율성, 수명 제한, 높은 유지보수 비용 등의 문제점을 AI의 지능으로 해결할 뿐만 아니라,사용자 중심의 전인적 지능형 빛 환경을 제공하며, 에너지 자립 및 넷 포지티브 달성을 통해 지속가능한 미래 사회를 구축하고, 극한 환경 적응성 및 다양한 산업 분야로의 혁신적 확장 가능성을 통해 인류 삶의 질 향상에 대단히 기여한다. The present invention relates to an AI-based intelligent photonic lighting platform that fundamentally shifts the paradigm of the lighting industry by comprehensively managing and optimizing the characteristics and energy of light sources through artificial intelligence (AI)-based autonomous learning and prediction. The invention comprises an AI control module that derives a lighting environment and energy management strategy optimized for the human body and environment, a light source module that precisely controls light at the nanophoton level, an intelligent complex energy generation and circulation management system, and a hyper-connected platform architecture that integrates sensors and communication functions and boasts high efficiency, long life, and a modular structure under AI-based integrated management. The invention not only resolves the problems of existing lighting, such as heat generation, inefficiency, limited lifespan, and high maintenance costs, through the intelligence of AI, but also provides a user-centered, holistic intelligent lighting environment, builds a sustainable future society by achieving energy independence and net positivity, and contributes significantly to the improvement of the quality of human life through adaptability to extreme environments and innovative expandability into various industrial fields.","assignee":"구교선; 구현우","inventors":["구교선","구현우"],"publication_date":"2025-11-25","filing_date":"2025-11-05","priority_date":"2025-11-05","cpc_codes":["H","H05","H05B","H05B47/00","H05B47/10","H05B47/105","G","G06","G06N","G06N20/00","G","G16","G16H","G16H20/00","G16H20/70","G","G16","G16H","G16H50/00","G16H50/70","H","H02","H02J","H02J50/00","H02J50/20","H","H04","H04B","H04B10/00","H04B10/11","H04B10/114","H04B10/116","H","H05","H05B","H05B45/00","H05B45/10","H","H05","H05B","H05B45/00","H05B45/50","H05B45/56","H","H05","H05B","H05B47/00","H05B47/10","H05B47/165","H","H05","H05B","H05B47/00","H05B47/20","Y","Y02","Y02B","Y02B20/00","Y02B20/40"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250164670A/en"},{"publication_number":"KR20250164671A","title":"AI-based Carbon-Sequestering Biomass Power Generation System and its Operating Method","abstract":"본 발명은 목질계 폐자원(재선충 피해목, 산불피해목, 홍수 피해목, 폐목재, 일반벌목재등)을 가스화하여 목탄가스를 생성하고, 이를 연소하여 전력을 생산하는 바이오매스 발전 시스템에 관한 것이다. 본 시스템은 발전 과정에서 발생되는 CO 2 를 Ca(OH) 2 수용액이 저장된 반응수조 하부에 기포상 주입하고 화학반응을 통해 고체 탄산칼슘(CaCO 3 )으로 전환하여 외부로의 연기 및 CO 2 배출 없이 현장에서 영구 고체 저장이 가능하도록 구성된다. 또한 압력·온도·유량·전압·전류 등 복수 센서 데이터를 AI 기반 컨트롤러가 통합 분석하여 연료공급량, 산소공급량 등을 실시간 제어함으로써 가스화 및 발전효율을 최적화한다. 상기 구성 전체는 모듈형 컨테이너 구조로 일체화되어 20kW, 100kW, 500kW 단위로 확장이 가능하다. The present invention relates to a biomass power generation system that gasifies wood waste resources (wood damaged by pine nematodes, wood damaged by forest fires, wood damaged by floods, waste wood, general logging wood, etc.) to produce charcoal gas, and combusts the charcoal gas to produce electricity. The system is configured to inject CO2 generated during the power generation process in the form of bubbles into the lower part of a reaction tank storing a Ca(OH) 2 aqueous solution, and convert it into solid calcium carbonate ( CaCO3 ) through a chemical reaction, thereby enabling permanent solid storage on site without emitting smoke or CO2 to the outside. In addition, an AI-based controller integrates and analyzes multiple sensor data such as pressure, temperature, flow rate, voltage, and current, and controls fuel supply, oxygen supply, etc. in real time, thereby optimizing gasification and power generation efficiency. The entire configuration is integrated into a modular container structure, allowing expansion in units of 20 kW, 100 kW, and 500 kW.","assignee":"강보원","inventors":["강보원"],"publication_date":"2025-11-25","filing_date":"2025-11-05","priority_date":"2025-11-05","cpc_codes":["C","C10","C10J","C10J3/00","C10J3/72","C10J3/723","B","B01","B01D","B01D53/00","B01D53/34","B01D53/46","B01D53/62","B","B01","B01D","B01D53/00","B01D53/34","B01D53/74","B01D53/77","B01D53/78","C","C01","C01F","C01F11/00","C01F11/18","C01F11/182","C","C10","C10J","C10J3/00","C10J3/72","C10J3/82","G","G05","G05B","G05B13/00","G05B13/02","G05B13/0265","G","G05","G05B","G05B13/00","G05B13/02","G05B13/04","G05B13/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","B","B01","B01D","B01D2257/00","B01D2257/50","B01D2257/504","C","C10","C10J","C10J2200/00","C10J2200/31","C","C10","C10J","C10J2300/00","C10J2300/09","C10J2300/0913","C10J2300/0916","C","C10","C10J","C10J2300/00","C10J2300/16","C10J2300/1603","C10J2300/1612","C","C10","C10J","C10J2300/00","C10J2300/16","C10J2300/1671","Y","Y02","Y02C","Y02C20/00","Y02C20/40"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250164671A/en"},{"publication_number":"CN121009285A","title":"A method for intelligent content filtering based on specified theme scenarios","abstract":"本申请涉及一种基于指定主题场景下智能内容过滤的方法，实现在指定主题场景下对ASR输出文本进行无关内容的精确过滤。所述方法包括：对零样本分类模型进行优化，得到优化的零样本分类模型；采用优化的零样本分类模型对第一ASR输出文本进行主题分类，得到第一无关内容文本；构建动态词库，计算所述第一ASR输出文本与所述动态词库的相关性，根据相关性筛选得到第二无关内容文本；对所述第一无关内容文本和所述第二无关内容文本进行比对重复内容，得到第三无关内容文本；在所述第一ASR输出文本中剔除所述第三无关内容文本，得到过滤后的第一ASR输出文本。 This application relates to a method for intelligent content filtering based on a specified topic scenario, achieving accurate filtering of irrelevant content from ASR output text within a specified topic scenario. The method includes: optimizing a zero-shot classification model to obtain an optimized zero-shot classification model; using the optimized zero-shot classification model to classify a first ASR output text into a topic, obtaining first irrelevant content text; constructing a dynamic lexicon, calculating the correlation between the first ASR output text and the dynamic lexicon, and filtering second irrelevant content text based on the correlation; comparing the first and second irrelevant content texts for duplicate content, obtaining third irrelevant content text; and removing the third irrelevant content text from the first ASR output text to obtain the filtered first ASR output text.","assignee":"Guangzhou Yuechuangzhishu Information Technology Co ltd","inventors":["张为华","杨毅","周伟","梁丽珍","莫斯韦","史欣月","李袁丹"],"publication_date":"2025-11-25","filing_date":"2025-10-28","priority_date":"2025-10-28","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G06F16/353","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F40/00","G06F40/10","G06F40/194","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G06F40/216","G","G06","G06F","G06F40/00","G06F40/20","G06F40/237","G06F40/242","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04","G","G10","G10L","G10L15/00","G10L15/08","G10L15/18","G","G10","G10L","G10L15/00","G10L15/26"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121009285A/en"},{"publication_number":"CN121010104A","title":"A method and system for generating emergency response procedures driven by a large model.","abstract":"The invention discloses a large model driven emergency disposal flow generation method and system, wherein the method comprises the steps of preprocessing data of an emergency disposal case, conducting document blocking on the preprocessed emergency disposal case by using a father-son blocking method, constructing a knowledge base and a vector library, conducting keyword and vector retrieval based on a BM25 sparse retrieval method and a BGE dense retrieval method, conducting weighted fusion on retrieval results to obtain a mixed retrieval result, constructing an emergency disposal flow to generate a prompt word, inputting an emergency description text, the mixed retrieval result and the prompt word into a large language model to generate a disposal flow conforming to a preset specification, verifying the disposal flow, and storing the verified disposal flow as a new emergency disposal case into the knowledge base and the vector library.","assignee":"CETC 28 Research Institute","inventors":["黄颖","章迪","崔隽","朱伟","李圣龙","郭政杰","徐尧","焦淼"],"publication_date":"2025-11-25","filing_date":"2025-10-28","priority_date":"2025-10-28","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/3332","G06F16/3335","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3344","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N5/00","G06N5/02","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121010104A/en"},{"publication_number":"CN121009401A","title":"A self-optimizing ultrasonic flow monitoring method based on cloud-centric neural networks","abstract":"The invention provides a cloud center neural network-based self-optimization ultrasonic flow monitoring method which comprises the following substeps of S1, S2, data storage and characteristic supplementation are carried out by a cloud center, S3, training and optimization of a cloud center neural network model, S4, issuing an optimal neural network model and carrying out local flow calculation, S5, optimizing the model based on an error result, introducing a cloud center deep neural network to carry out flow prediction, uploading data to a cloud, eliminating the limitation of an ultrasonic flowmeter on the calculation capability of equipment, improving the calculation precision, improving the adaptability to complex flow state and working condition change, reducing measurement errors, carrying out data storage by the cloud center, associating with a global knowledge base, sharing data and experiences among different equipment, reducing artificial experience dependence, and solving the problem of equipment dependence on local experience.","assignee":"Maxtor Instrument Ltd By Share Ltd","inventors":["王伟","王志伟","陈志伟","邵文强"],"publication_date":"2025-11-25","filing_date":"2025-10-28","priority_date":"2025-10-28","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098","G","G06","G06F","G06F2123/00","G06F2123/02","G","G06","G06F","G06F2218/00","G06F2218/08","G","G06","G06F","G06F2218/00","G06F2218/12"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121009401A/en"},{"publication_number":"CN121012925A","title":"A compression method and device for image recognition networks","abstract":"The application discloses a compression method and equipment of an image recognition network, which relate to the technical field of artificial intelligence and comprise the steps of firstly carrying out pseudo quantization based on preset initial bits to obtain initial pseudo quantized weights, then dynamically determining target quantization bits of an adaptive target convolution layer by combining the initial weights through a preset mapping function, so that quantization precision can be matched with feature importance differences of different convolution layers, the problem that key channel information loss or insufficient quantization is caused by fixed quantization bits in the traditional method is solved, and the effects of reducing key feature loss in the quantization process and improving model precision after quantization are achieved. Furthermore, the scaling coefficient matrix is solved by combining the target pseudo-quantized weight and the batch normalized layer statistical parameters so as to reflect the quantization loss degree of each channel, and then the weight of the next adjacent convolution layer is subjected to targeted scaling, so that the cross-layer accurate compensation of the quantization loss is realized, and the problems that the interlayer relevance is ignored and the compensation effect is poor in the traditional compensation mode are solved.","assignee":"Inspur Electronic Information Industry Co Ltd","inventors":["尹文枫","董刚"],"publication_date":"2025-11-25","filing_date":"2025-10-28","priority_date":"2025-10-28","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06V","G06V10/00","G06V10/40","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/7715","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121012925A/en"},{"publication_number":"CN121009125A","title":"A Contract Review System and Method Based on Multimodal Analysis and Risk Early Warning","abstract":"本发明公开一种基于多模态解析与风险预警的合同评审系统及方法，涉及智能文档处理技术领域；包括：步骤1：进行文档解析，步骤2：进行要素提取：基于自然语言处理方法及预训练的大语言模型从解析后的文本信息中提取要素信息，步骤3：进行语义风险识别，步骤4：进行附件比对：针对用户上传的合同附件，提取合同附件中的要素信息，并与合同正文中的要素信息进行交叉比对，检查是否存在不一致或缺失的情况，步骤5：进行OA对接，步骤6：进行合同文本的脱敏处理；本发明聚焦于合同全生命周期中的评审环节，解决现有技术中合同审核效率低、风险识别滞后、系统间数割裂等问题。 This invention discloses a contract review system and method based on multimodal parsing and risk warning, belonging to the field of intelligent document processing technology; it includes: Step 1: document parsing; Step 2: element extraction: extracting element information from the parsed text information based on natural language processing methods and pre-trained large language models; Step 3: semantic risk identification; Step 4: attachment comparison: extracting element information from the contract attachments uploaded by users and cross-comparing it with the element information in the contract text to check for inconsistencies or omissions; Step 5: OA integration; Step 6: anonymizing the contract text; This invention focuses on the review stage in the entire contract lifecycle, solving problems such as low contract review efficiency, delayed risk identification, and data fragmentation between systems in existing technologies.","assignee":"Inspur Enterprise Cloud Technology Shandong Co ltd","inventors":["徐航","刘慧欣","任寿杰"],"publication_date":"2025-11-25","filing_date":"2025-10-28","priority_date":"2025-10-28","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2455","G06F16/24564","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/242","G06F16/243","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2452","G06F16/24522","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/951","G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6227","G","G06","G06F","G06F40/00","G06F40/10","G06F40/194","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/103","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/18"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121009125A/en"},{"publication_number":"CN121009809A","title":"Model-based collaborative design aid method and system for spinning equipment","abstract":"The invention belongs to the technical field of computer aided design, and provides a spinning equipment design aided method and system based on model collaboration. The method comprises the steps of constructing a collaborative design system, enabling a large language model to receive user design requirements, performing task decomposition, calling corresponding professional agency models to execute design subtasks, enabling each professional agency model to feed simulation calculation results back to a closed-loop feedback engine, enabling the closed-loop feedback engine to analyze feedback results of each professional agency model and generate attention prompt signals pointing to potential design conflicts or optimization bottlenecks based on cross-disciplinary parameter coupling relations, enabling the large language model to perform multi-disciplinary coupling analysis by utilizing the attention prompt signals, focusing on key parameters and key disciplines pointed by the attention prompt signals, dynamically adjusting task planning and design parameters, and driving related professional agency models to perform directed iterative optimization. The invention can effectively avoid the blind iteration problem, so that the convergence speed and the scheme quality of the design are improved with high quality.","assignee":"Kuche Lihua Textile Co ltd; Donghua University","inventors":["黄克华","黄奕菲","贾春胜","鲍劲松","郭长华","何俊飞","崔晨雨"],"publication_date":"2025-11-25","filing_date":"2025-10-28","priority_date":"2025-10-28","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06F","G06F30/00","G06F30/10","G06F30/17","G","G06","G06N","G06N3/00","G06N3/004","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","Y","Y02","Y02P","Y02P90/00","Y02P90/30"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN121009809A/en"},{"publication_number":"CN120997454A","title":"Lightweight methods, systems, devices and storage media for 3D plant models","abstract":"本申请提供一种植物三维模型轻量化方法、系统、设备及存储介质，涉及模型处理技术领域。该方法包括：基于结构特征将植物三维模型划分为多个组成部分；基于第一减面方式对所述第一组成部分进行减面操作，得到第一减面部分；其中，所述第一组成部分为任一所述组成部分，所述第一减面方式是基于第一组成部分的结构特征确定的；基于多个所述第一减面部分生成轻量化模型。该方法可以通过针对植物三维模型的不同组成部分采用适应的轻量化策略进行模型减面，对模型进行多层次的优化处理，可以提高减小模型文件大小的能力，可以提高存储植物三维模型以及提高植物三维模型数据传输的效率，从而提升模型使用的流畅性。 This application provides a method, system, device, and storage medium for lightweighting plant 3D models, relating to the field of model processing technology. The method includes: dividing a plant 3D model into multiple components based on structural features; performing a surface reduction operation on the first component using a first surface reduction method to obtain a first surface-reduced portion; wherein the first component is any of the aforementioned components, and the first surface reduction method is determined based on the structural features of the first component; and generating a lightweight model based on multiple first surface-reduced portions. This method can reduce the surface area of the model by employing adaptive lightweighting strategies for different components of the plant 3D model, performing multi-level optimization processing on the model, improving the ability to reduce model file size, improving the efficiency of storing plant 3D models, and improving the efficiency of plant 3D model data transmission, thereby enhancing the smoothness of model usage.","assignee":"China Southwest Architectural Design and Research Institute Co Ltd","inventors":["姜卓","张�成","梁逍","张静","孙浩","高飞","朱驰浩"],"publication_date":"2025-11-21","filing_date":"2025-10-27","priority_date":"2025-10-27","cpc_codes":["G","G06","G06T","G06T17/00","G06T17/20","G06T17/205","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T15/00","G06T15/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120997454A/en"},{"publication_number":"CN120994656A","title":"A dynamic data quality rule intelligent generation and adaptive correction system","abstract":"The invention provides an intelligent generation and self-adaptive correction system for dynamic data quality rules, which belongs to the field of data management and data management. By utilizing the powerful natural language understanding, code generation and logic reasoning capabilities of the large model, the intelligent generation, dynamic evaluation and self-adaptive correction of the data quality rules are realized, a self-evolution and continuous optimized data quality control closed loop is formed, and the efficiency and accuracy of data management are remarkably improved.","assignee":"Inspur Software Technology Co Ltd","inventors":["�田�浩","王彦功","李可君","陈焕新","寇智铭","李娇","袁富强","刘金革"],"publication_date":"2025-11-21","filing_date":"2025-10-27","priority_date":"2025-10-27","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/21","G06F16/215","G","G06","G06F","G06F16/00","G06F16/20","G06F16/23","G06F16/2379","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2455","G06F16/24564","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120994656A/en"},{"publication_number":"CN120996216A","title":"Reasoning methods, systems, devices, and media for dynamic routing hybrid expert models","abstract":"本发明公开了一种动态路由混合专家模型的推理方法、系统、设备及介质，它们是相对应的方案，方案中：通过自动化的并行策略搜索，本发明能将模型切分为运行时间均衡的流水线阶段，减少计算单元的空闲率，进而有效地提升计算单元的执行效率；并且，本发明的自动化搜索过程通过细粒度的并行策略划分，能够在满足内存限制的情况下尽可能降低最大的流水线阶段运行时间，从而提高模型的推理性能。 This invention discloses an inference method, system, device, and medium for a dynamic routing hybrid expert model. These are corresponding solutions. In these solutions, through automated parallel strategy search, the invention can divide the model into pipeline stages with balanced runtime, reducing the idle rate of computing units and thus effectively improving the execution efficiency of computing units. Furthermore, the automated search process of this invention, through fine-grained parallel strategy partitioning, can minimize the runtime of the largest pipeline stage while meeting memory constraints, thereby improving the inference performance of the model.","assignee":"University of Science and Technology of China USTC","inventors":["白有辉","傅申","金泽文","唐承捷","方驰正","李�诚"],"publication_date":"2025-11-21","filing_date":"2025-10-27","priority_date":"2025-10-27","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/04","G06N5/043","G","G06","G06N","G06N5/00","G06N5/01"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120996216A/en"},{"publication_number":"CN120998294A","title":"A method for predicting drug-target interaction associations based on joint network attention","abstract":"The invention belongs to the field of bioinformatics, and relates to a drug and target action association prediction method based on joint network attention. Firstly, extracting context information of a medicine substructure and a protein residue through multi-task self-supervision feature learning of a medicine molecular diagram and a protein sequence, so as to obtain high-precision representation features, secondly, combining a machine learning model to realize interaction relation between a medicine and a target, and finally, realizing action correlation prediction of the medicine and the target through a trained model. The invention can maintain excellent prediction performance and strengthening capability under the condition of limited marked data. Experimental results show that the method has remarkable advantages in aspects of drug discovery, target spot screening, candidate drug action mechanism identification and the like, and can provide efficient and reliable tool support for new drug research and development.","assignee":"Ludong University","inventors":["李锦龙","周树森","刘通","柳婵娟","王庆军","臧睦君"],"publication_date":"2025-11-21","filing_date":"2025-10-27","priority_date":"2025-10-27","cpc_codes":["G","G16","G16B","G16B15/00","G16B15/30","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06F","G06F18/00","G06F18/20","G06F18/27","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G16","G16B","G16B40/00","G","G16","G16C","G16C20/00","G16C20/50","G","G16","G16C","G16C20/00","G16C20/70"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120998294A/en"},{"publication_number":"CN120997886A","title":"A Visual-Text Cross-Modal Giant Panda Behavior Recognition Method Based on Attention Mechanism","abstract":"The invention provides a visual-text cross-modal panda behavior recognition method based on an attention mechanism, which relates to the technical field of the attention mechanism, and comprises the steps of inputting a multi-modal dataset into an initial model, extracting panda behavior characteristics, introducing a customized cross-modal attention mechanism to realize characteristic alignment, strengthening the interaction depth of visual and text characteristics, breaking through the bottleneck of insufficient semantic fusion, constructing a video frame sequence-text description pair sample based on the characteristics, developing bidirectional matching learning in a unified embedding space through a cross-modal characterization network, losing optimization parameters by symmetrical cross entropy, combining verification and early stop curing models, capturing video time sequence information, solving the problem of incomplete behavior dynamic characterization, finally optimizing a visual part of target panda behavior data, adjusting a preprocessing strategy according to quality parameters, enabling an optimization result to assist in focusing key characteristics of the attention mechanism, improving the current situation that preprocessing is not unified standard and cannot be fed back, and finally accurately recognizing a target behavior category.","assignee":"CHENGDU RESEARCH BASE OF GIANT PANDA BREEDING","inventors":["陈鹏","郑维超","金子成","张萍","侯蓉","何梦楠","马莹","罗概","吴鹏程","巫林","杨淑曼"],"publication_date":"2025-11-21","filing_date":"2025-10-27","priority_date":"2025-10-27","cpc_codes":["G","G06","G06V","G06V40/00","G06V40/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/40","G06V20/46","G","G06","G06V","G06V30/00","G06V30/10","G06V30/18","G06V30/1801","G06V30/18019","G06V30/18038","G06V30/18048","G06V30/18057","G","G06","G06V","G06V30/00","G06V30/10","G06V30/19","G06V30/19007","G","G06","G06V","G06V30/00","G06V30/10","G06V30/19","G06V30/19007","G06V30/19093"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120997886A/en"},{"publication_number":"CN120995184A","title":"An industrial fault detection method and system based on dynamic drift sensing and diffusion enhancement","abstract":"本发明涉及故障检测技术领域，尤其是涉及一种基于动态漂移感知与扩散增强的工业故障检测方法及系统。方法包括构建无监督故障检测DDA‑DE模型，其中，利用动态漂移感知DDA处理工业数据流以建立统计分布基线和漂移阈值；利用扩散增强异常检测DE确定初始模型参数与异常阈值；基于扩散策略对无监督故障检测模型进行概念漂移的鲁棒性增强；本发明所提出的基于工业增强的马氏距离实时漂移感知算法能够更高效检测出概念漂移现象，漂移感知算法和故障检测分类器并行的设计实现数据漂移与异常事件的协同检测。 This invention relates to the field of fault detection technology, and in particular to an industrial fault detection method and system based on dynamic drift sensing and diffusion enhancement. The method includes constructing an unsupervised fault detection DDA-DE model, wherein dynamic drift sensing DDA is used to process industrial data streams to establish a statistical distribution baseline and drift threshold; diffusion-enhanced anomaly detection DE is used to determine initial model parameters and anomaly thresholds; the robustness of the unsupervised fault detection model to concept drift is enhanced based on a diffusion strategy; the proposed industrial-enhanced Mahalanobis distance real-time drift sensing algorithm can more efficiently detect concept drift phenomena, and the parallel design of the drift sensing algorithm and the fault detection classifier achieves collaborative detection of data drift and anomaly events.","assignee":"Yantai University","inventors":["单垚","刘兆伟","杨利成","王莹洁","宋永超","姜岸佐","王鹏","侯文涵"],"publication_date":"2025-11-21","filing_date":"2025-10-27","priority_date":"2025-10-27","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120995184A/en"},{"publication_number":"CN120996123A","title":"Model update method, apparatus, computer equipment and storage medium","abstract":"The embodiment of the application discloses a model updating method, a model updating device, computer equipment and a storage medium. The method comprises the steps of determining computing capability scores of all computing nodes in a plurality of computing nodes according to computing power values and memory values corresponding to the computing nodes, screening out a plurality of target computing nodes in the plurality of computing nodes according to the computing capability scores, obtaining data to be processed, dividing batch processing data of all target computing nodes in the data to be processed according to the computing power values and the memory values of the computing nodes, obtaining model parameter matrixes which are output after the target computing nodes load target models to process the corresponding batch processing data, determining updated model parameter matrixes of the target models according to the model parameter matrixes, determining differential model parameter matrixes between the updated model parameter matrixes and the model parameter matrixes of the target models, and updating the target models in the computing nodes according to the differential model parameter matrixes to obtain updated target models.","assignee":"Peng Cheng Laboratory","inventors":["李若南","顾钊铨","李金龙","刘劼","宗瑞","王海燕","邵迪","吴世彬"],"publication_date":"2025-11-21","filing_date":"2025-10-27","priority_date":"2025-10-27","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120996123A/en"},{"publication_number":"CN120994531A","title":"Chatbot Intelligent Debugging System and Method","abstract":"The invention provides an intelligent debugging system and method for a chat robot, which relate to the technical field of artificial intelligence and natural language processing, wherein the intelligent debugging system for the chat robot further integrates a predictive prompt engine, a natural language understanding service, an enhanced root cause analysis service and the like on the basis of providing a visual flow construction and an integrated test debugging interface, and the engines/services, the flow service, the dialogue management service and the like work cooperatively, thereby providing more intelligent, deeper and more comprehensive support in the whole life cycle of the chat robot in design, test and debugging, obviously improving the development quality and efficiency of the chat robot and reducing the technical threshold.","assignee":"Beisen Cloud Computing Co ltd","inventors":["尤朝阳","纪伟国","孙江"],"publication_date":"2025-11-21","filing_date":"2025-10-27","priority_date":"2025-10-27","cpc_codes":["G","G06","G06F","G06F11/00","G06F11/36","G06F11/362","G06F11/366","G","G06","G06F","G06F11/00","G06F11/36","G06F11/362","G06F11/3644","G","G06","G06F","G06F11/00","G06F11/36","G06F11/3698","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120994531A/en"},{"publication_number":"CN120997223A","title":"A Road Defect Detection Method Based on RDD-YOLOv8 Model","abstract":"本发明公开了一种基于RDD‑YOLOv8模型的道路缺陷检测方法，首先构建道路缺陷数据集；然后构建道路缺陷检测算法模型，以YOLOv8n为基础算法，设计了RD注意力机制，并将其融入YOLOv8n算法主干网络中，提出了新型损失函数FEAIOU，并使用预处理后的数据集对改进后的YOLO算法模型进行训练，采取mAP、Params和FLOPs指标评价算法模型，得到最佳算法模型；最后，将改进后算法模型进行边缘部署，将待检测道路缺陷图像或视频送入训练好的算法模型中，输出道路缺陷类别。本申请通过RDD‑YOLOv8模型，增强了对复杂缺陷的识别性能和预测分类能力，能够提高检测精度和效率。 This invention discloses a road defect detection method based on the RDD-YOLOv8 model. First, a road defect dataset is constructed. Then, a road defect detection algorithm model is built, using YOLOv8n as the base algorithm. An RD attention mechanism is designed and integrated into the YOLOv8n algorithm backbone network. A novel loss function, FEAIOU, is proposed. The improved YOLO algorithm model is trained using a preprocessed dataset. The algorithm model is evaluated using mAP, Params, and FLOPs metrics to obtain the optimal algorithm model. Finally, the improved algorithm model is deployed at the edge, and images or videos of road defects to be detected are fed into the trained algorithm model to output the road defect category. This application enhances the recognition performance and predictive classification ability for complex defects through the RDD-YOLOv8 model, thereby improving detection accuracy and efficiency.","assignee":"NANJING SPECIAL EQUIPMENT INSPECTION INSTITUTE","inventors":["孙勇","刘润捷","路成龙","庆光蔚","刘堃","倪大进","王爽"],"publication_date":"2025-11-21","filing_date":"2025-10-27","priority_date":"2025-10-27","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/0002","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06T","G06T7/00","G06T7/70","G06T7/73","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120997223A/en"},{"publication_number":"CN120995362A","title":"Multifunctional Integrated Intelligent Monitoring Method and Device for Near-Power Construction","abstract":"本发明公开了一种近电施工多功能一体化智能监控方法及装置，其中方法包括：基于激光雷达、摄像头和电场传感器采集近电施工历史数据；对所采集的近电施工历史数据进行预处理，并基于预设的数据融合方法将预处理后数据整合为多源异构数据；基于多源异构数据构建数据集；基于卷积神经网络CNN构建施工安全检测模型；采用数据集训练施工安全检测模型；实时采集近电施工数据，对数据进行预处理并将预处理后数据整合为多源异构数据，将多源异构数据输入训练好的施工安全检测模型，实时检测施工安全隐患及施工异常。本发明具有安全性强、实时性强准确性高等突出优势。 This invention discloses a multi-functional integrated intelligent monitoring method and device for near-electric construction. The method includes: collecting historical data on near-electric construction based on lidar, cameras, and electric field sensors; preprocessing the collected historical data and integrating the preprocessed data into multi-source heterogeneous data based on a preset data fusion method; constructing a dataset based on the multi-source heterogeneous data; constructing a construction safety detection model based on a convolutional neural network (CNN); training the construction safety detection model using the dataset; collecting near-electric construction data in real time, preprocessing the data, integrating the preprocessed data into multi-source heterogeneous data, inputting the multi-source heterogeneous data into the trained construction safety detection model, and detecting construction safety hazards and anomalies in real time. This invention has significant advantages such as strong security, strong real-time performance, and high accuracy.","assignee":"State Grid Electric Power Engineering Research Institute Co ltd; State Grid Shanghai Electric Power Co Ltd","inventors":["周亚傲","倪金禄","江明","周耀俊","朱纯","李洋"],"publication_date":"2025-11-21","filing_date":"2025-10-27","priority_date":"2025-10-27","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2433","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120995362A/en"},{"publication_number":"KR20250163285A","title":"Method, program, and apparatus for diagnosing heart disease based on ecg signal information","abstract":"본 개시는 심전도 신호에 기초하여 심장 질환을 진단하는 방법, 프로그램 및 장치를 제공한다. 본 개시의 일 실시 예에 따른 심전도 신호에 기초하여 심장 질환을 진단하는 방법은 사용자에 대한 심전도 신호를 획득하는 단계와 상기 획득된 심전도 신호에 기초하여, 상기 사용자의 심장 질환을 진단하는 단계를 포함하고, 상기 진단하는 단계는, 상기 획득된 심전도 신호를 기 학습된 신경망 모델에 입력하여, 상기 사용자의 심장 질환에 대응하는 제1 스코어를 산출하는 단계, 상기 산출된 제1 스코어에 기초하여 상기 사용자의 심장 질환에 대한 추가 진단 여부를 결정하여 상기 사용자에 대한 심전도 신호의 재 획득하는 단계와 상기 재 획득된 심전도 신호에 기초하여 상기 사용자의 심장 질환을 추가 진단하는 단계를 포함한다. The present disclosure provides a method, a program, and a device for diagnosing a heart disease based on an electrocardiogram signal. According to one embodiment of the present disclosure, a method for diagnosing a heart disease based on an electrocardiogram signal includes the steps of: acquiring an electrocardiogram signal for a user; and diagnosing a heart disease of the user based on the acquired electrocardiogram signal, wherein the diagnosing step includes the steps of: inputting the acquired electrocardiogram signal into a pre-trained neural network model to calculate a first score corresponding to the heart disease of the user; determining whether to further diagnose the heart disease of the user based on the calculated first score, thereby re-acquiring the electrocardiogram signal for the user; and further diagnosing the heart disease of the user based on the re-acquired electrocardiogram signal.","assignee":"주식회사 메디컬에이아이","inventors":["권준명"],"publication_date":"2025-11-20","filing_date":"2025-11-12","priority_date":"2023-05-23","cpc_codes":["A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/318","A61B5/346","A","A61","A61B","A61B5/00","A","A61","A61B","A61B5/00","A61B5/103","A61B5/11","A61B5/1118","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/318","A61B5/332","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7275","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H40/00","G16H40/60","G16H40/63","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/30"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250163285A/en"},{"publication_number":"AU2025263751A1","title":"1 system and method for customer journey event representation learning and outcome prediction using neural sequence models","abstract":"1 A system and method are presented for customer journey event representation learning and outcome prediction using neural sequence models. A plurality of events are input into a module where each event has a schema comprising characteristics of the events and their modalities (web clicks, calls, emails, chats, etc.). The events of different modalities can be captured using different schemas and therefore embodiments described herein are schema-agnostic. Each event is represented as a vector of some number of numbers by the module with a plurality of vectors being generated in total for each customer visit. The vectors are then used in sequence learning to predict real-time next best actions or outcome probabilities in a customer journey using machine learning algorithms such as recurrent neural networks.","assignee":"Genesys Cloud Services Inc","inventors":["Maciej Dabrowski","Aravind GANAPATHIRAJU","Emir MUNOZ","Sapna NEGI","Veera Elluru RAGHAVENDRA","Felix Immanuel Wyss"],"publication_date":"2025-11-20","filing_date":"2025-11-04","priority_date":"2019-04-09","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q30/00","G06Q30/01","G06Q30/015","G06Q30/016","G","G06","G06F","G06F21/00","G06F21/60","G06F21/602","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0281","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0631","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N7/00","G06N7/01"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025263751A1/en"},{"publication_number":"AU2025263749A1","title":"Digital therapeutic systems and methods","abstract":"74 Methods and devices include identifying a plurality of target users for the digital therapeutic based on one or more target parameters, conducting outreach to one or more of the plurality of target users using an outreach medium, identifying an activation mechanism to optimize use of the digital therapeutic, and encouraging an engagement level of the digital therapeutic by one or more of the plurality of target users.","assignee":"WellDoc Inc","inventors":["Carey HUTCHINS","Anand Iyer","Malinda Peeples","Vinayak SHENOY"],"publication_date":"2025-11-20","filing_date":"2025-11-04","priority_date":"2019-12-04","cpc_codes":["G","G16","G16H","G16H40/00","G16H40/60","G16H40/67","G","G06","G06N","G06N20/00","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H15/00","G","G16","G16H","G16H20/00","G","G16","G16H","G16H20/00","G16H20/10","G","G16","G16H","G16H20/00","G16H20/30","G","G16","G16H","G16H20/00","G16H20/60","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H70/00","G16H70/20","G","G16","G16H","G16H80/00"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025263749A1/en"},{"publication_number":"AU2025260002A1","title":"Deciphering of detected silent speech","abstract":"1006232449 A method for synthesizing speech, the method comprising: receiving input signals from a human subject that are indicative of intended speech by the human subject; analyzing the signals to extract words corresponding to the intended speech, such that in at least some time intervals of the 5 intended speech, multiple candidate phonemes are extracted together with respective probabilities that each of the candidate phoneme corresponds to the intended speech in a given time interval; and synthesizing audible speech responsively to the extracted phonemes, such that in the at least some of the time intervals, the audible speech is synthesized by mixing the multiple candidate phonemes responsively to the respective probabilities. 10","assignee":"Q Cue Ltd","inventors":["Avi BARLIYA","Doron Gazit","Giora Kornblau","Aviad Maizels","Yonatan Wexler"],"publication_date":"2025-11-20","filing_date":"2025-11-03","priority_date":"2021-08-04","cpc_codes":["G","G10","G10L","G10L15/00","G10L15/24","G10L15/25","A","A61","A61B","A61B5/00","A61B5/0059","A61B5/0077","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4803","A","A61","A61B","A61B5/00","A61B5/68","A61B5/6801","A61B5/6802","A61B5/6803","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06V","G06V10/00","G06V10/10","G06V10/12","G06V10/14","G06V10/147","G","G06","G06V","G06V10/00","G06V10/20","G06V10/24","G06V10/245","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G06V40/168","G06V40/171","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G06V40/174","G","G10","G10L","G10L13/00","G","G10","G10L","G10L13/00","G10L13/02","G10L13/027","G","G10","G10L","G10L15/00","G10L15/24","A","A61","A61B","A61B2562/00","A61B2562/02","A61B2562/0233","A","A61","A61B","A61B2576/00","A61B2576/02","A","A61","A61B","A61B5/00","A61B5/0033","A61B5/004","A","A61","A61B","A61B5/00","A61B5/103","A61B5/11","A61B5/1113","A61B5/1114","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","A","A61"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025260002A1/en"},{"publication_number":"KR20250163271A","title":"Interactive system for assisting user to use medium","abstract":"본 개시는 사용자의 미디엄 이용을 보조하기 위한 상호작용 시스템에 관한 것이다. 보다 구체적으로 사용자가 미디엄을 읽거나 미디엄을 통해 정해진 액티비티(activity)를 수행할 때, 미디엄의 내용과 관련된 사용자의 궁금증을 해소시키거나 미디엄에 대한 사용자의 이해를 돕거나 사용자가 액티비티를 원활하게 수행할 수 있도록 보조하는 상호작용 시스템에 관한 것이다. The present disclosure relates to an interactive system that assists users in using a medium. More specifically, the disclosure relates to an interactive system that, when a user reads a medium or performs an activity specified through the medium, resolves the user's curiosity related to the medium's content, aids the user's understanding of the medium, or assists the user in smoothly performing the activity.","assignee":"주식회사 네오랩컨버전스","inventors":["이상규","국수열"],"publication_date":"2025-11-20","filing_date":"2025-11-03","priority_date":"2024-02-25","cpc_codes":["G","G06","G06F","G06F3/00","G06F3/16","G06F3/167","G","G06","G06F","G06F3/00","G","G06","G06F","G06F3/00","G06F3/002","G06F3/005","G","G06","G06F","G06F3/00","G06F3/01","G06F3/03","G06F3/033","G06F3/0354","G","G06","G06F","G06F3/00","G06F3/01","G06F3/03","G06F3/033","G06F3/0354","G06F3/03545","G","G06","G06F","G06F3/00","G06F3/16","G","G06","G06K","G06K19/00","G06K19/06","G","G06","G06K","G06K19/00","G06K19/06","G06K19/06009","G06K19/06037","G","G06","G06N","G06N20/00","G","G10","G10L","G10L13/00","G10L13/08","G","G10","G10L","G10L15/00","G10L15/26","H","H04","H04R","H04R1/00","H04R1/08","H","H04","H04R","H04R1/00","H04R1/10","H","H04","H04W","H04W4/00","H04W4/80"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250163271A/en"},{"publication_number":"CO2025015612A2","title":"Placement of distributed unit network functions (du nf) in a network","abstract":"Se describe un procedimiento, un sistema y un aparato. Se describe un primer nodo de red (NN) configurado para comunicarse con una pluralidad de NN en una red. El primer nodo de red está configurado para, y/o incluye una interfaz de comunicación y/o circuitos de procesamiento configurados para determinar una pluralidad de funciones de red de unidad distribuida (DU NF) que serán alojadas por un grupo de NN de la pluralidad de NN basándose en un objetivo de retardo de cada dispositivo inalámbrico (WD) de una pluralidad de WD, donde cada DU NF está asociada a al menos una portadora componente utilizable por al menos un WD, y la pluralidad de DU NF se determina utilizando un procedimiento de aprendizaje. Además, cada NN del grupo de NN se desencadena para alojar una NF DU correspondiente de la pluralidad de DU. A procedure, a system, and an apparatus are described. A first network node (NN) configured to communicate with a plurality of NNs in a network is described. The first network node is configured to, and/or includes a communication interface and/or processing circuitry configured to determine a plurality of distributed unit network functions (DU NFs) to be hosted by a group of NNs from the plurality of NNs based on a delay target for each wireless device (WD) from a plurality of WDs, where each DU NF is associated with at least one component carrier usable by at least one WD, and the plurality of DU NFs is determined using a learning procedure. Furthermore, each NN in the group of NNs is triggered to host a corresponding DU NF from the plurality of DUs.","assignee":"Ericsson Telefon Ab L M","inventors":["Roghayeh Joda","Sima Naseri","Mona Hashemi","Christopher Richards"],"publication_date":"2025-11-19","filing_date":"2025-11-10","priority_date":"2023-04-14","cpc_codes":["H","H04","H04W","H04W24/00","H04W24/02","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","H","H04","H04L","H04L41/00","H04L41/14","H04L41/145","H","H04","H04L","H04L41/00","H04L41/16","H","H04","H04L","H04L43/00","H04L43/08","H04L43/0852","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","H","H04","H04W","H04W88/00","H04W88/08","H04W88/085"],"country":"CO","kind":"application","source_url":"https://patents.google.com/patent/CO2025015612A2/en"},{"publication_number":"KR20250162480A","title":"Vacuum pump management apparatus and vacuum pump management method using the same","abstract":"진공펌프 관리 시스템과 이를 이용한 진공펌프 관리 방법이 개시된다. 본 발명의 일 실시예에 따른 진공펌프 관리 방법은, 챔버 내부를 대기압에서 저진공 상태로 조성하는 제1 진공펌프와, 상기 제1 진공펌프에 의해 미리 설정된 크로스오버 압력에 도달한 후 동작을 개시하여 상기 챔버 내부를 고진공 상태로 조성하고, 상기 제1 진공펌프와 서로 다른 진공 용량을 가지며 순차적으로 동작하는 제2 진공펌프를 포함하는, 복수개의 진공펌프에 각각 데이터가 발생하는 데이터 발생 단계; 상기 제1 진공펌프 및 상기 제2 진공펌프를 포함하는 진공펌프 단위로 매칭되고, 엣지 컴퓨팅에 기초하여 동작하는 노드가 상기 데이터를 수신하는 데이터 전송 단계; 상기 노드에서 상기 데이터에 기초해, 상기 제1 진공펌프의 사용연한 증대를 목적으로, 상기 크로스오버 압력을 상향 조정하여 상기 제2 진공펌프의 동작 개시 시간을 앞당기도록 하는 동작 신호를 상기 진공펌프에 전송하는 신호 전송 단계; 및 상기 동작 신호 기초해, 상기 제1 진공펌프의 자체 부하를 낮추고 상기 제2 진공펌프를 조기 가동시키는 진공펌프 동작 단계를 포함할 수 있다. 본 발명의 다른 실시예에 따른 진공펌프 관리 시스템은, 챔버 내부를 대기압에서 저진공 상태로 조성하는 제1 진공펌프와, 상기 제1 진공펌프에 의해 미리 설정된 크로스오버 압력에 도달한 후 동작을 개시하여 상기 챔버 내부를 고진공 상태로 조성하고, 상기 제1 진공펌프와 서로 다른 진공 용량을 가지며 순차적으로 동작하는 제2 진공펌프를 포함하는 복수개의 진공펌프; 상기 진공펌프에 각각 구비되어, 상기 진공펌프의 데이터를 검출하는 복수개의 센서; 및 상기 센서로부터 수신한 데이터를 처리하고, 상기 데이터에 기반한 신호를 생성해 상기 진공펌프의 동작을 제어하는 노드를 포함하고, 상기 노드는, 상기 제1 진공펌프 및 상기 제2 진공펌프를 포함하는 진공펌프 단위로 매칭되며, 엣지 컴퓨팅에 기초해 하나의 동작 단위로 상기 복수개의 진공펌프들을 연계하여 제거 가능하고, 상기 제1 진공펌프의 사용연한 증대를 목적으로, 상기 크로스오버 압력을 상향 조정하여 상기 제2 진공펌프의 동작 개시 시간을 앞당기도록 제어 가능할 수 있다. A vacuum pump management system and a vacuum pump management method using the same are disclosed. A vacuum pump management method according to one embodiment of the present invention may include a data generation step in which data is generated for each of a plurality of vacuum pumps, including a first vacuum pump for creating a low vacuum state inside a chamber from atmospheric pressure, and a second vacuum pump for creating a high vacuum state inside the chamber by starting operation after a preset crossover pressure is reached by the first vacuum pump and sequentially operating; a data transmission step in which a node that is matched with a vacuum pump unit including the first vacuum pump and the second vacuum pump and operates based on edge computing receives the data; a signal transmission step in which the node transmits an operation signal to the vacuum pump based on the data, for increasing the crossover pressure to advance an operation start time of the second vacuum pump for the purpose of extending the service life of the first vacuum pump; and a vacuum pump operation step in which, based on the operation signal, a self-load of the first vacuum pump is lowered and the second vacuum pump is operated early. According to another embodiment of the present invention, a vacuum pump management system includes a plurality of vacuum pumps, including a first vacuum pump for creating a low vacuum state inside a chamber from atmospheric pressure, and a second vacuum pump for creating a high vacuum state inside the chamber by starting operation after a preset crossover pressure is reached by the first vacuum pump and having a vacuum capacity different from that of the first vacuum pump and operating sequentially; a plurality of sensors each provided in the vacuum pumps for detecting data of the vacuum pumps; and a node for processing data received from the sensors and generating a signal based on the data to control the operation of the vacuum pumps, wherein the node is matched to a vacuum pump unit including the first vacuum pump and the second vacuum pump, and can be linked and removed as a single operation unit based on edge computing, and can be controlled to advance an operation start time of the second vacuum pump by upwardly adjusting the crossover pressure for the purpose of extending the service life of the first vacuum pump.","assignee":"엔셀 주식회사","inventors":["임용일"],"publication_date":"2025-11-18","filing_date":"2025-11-11","priority_date":"2021-07-14","cpc_codes":["F","F04","F04B","F04B37/00","F04B37/10","F04B37/14","F","F04","F04B","F04B41/00","F04B41/06","F","F04","F04B","F04B49/00","F04B49/007","F","F04","F04B","F04B49/00","F04B49/06","F","F04","F04B","F04B49/00","F04B49/10","F","F04","F04B","F04B51/00","F","F04","F04C","F04C25/00","F04C25/02","F","F04","F04C","F04C28/00","F04C28/28","F","F04","F04D","F04D19/00","F04D19/02","F04D19/04","F","F04","F04D","F04D27/00","F04D27/001","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0259","G05B23/0283","G","G06","G06N","G06N20/00","G","G08","G08B","G08B21/00","G08B21/18","G08B21/187","F","F04","F04B","F04B2207/00","F04B2207/70"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250162480A/en"},{"publication_number":"KR20250162477A","title":"Method, program, and device for providing artificial intelligence-based medical history summarization service","abstract":"서버의 프로세서에 의해 실행되는 인공지능 기반 문진 요약 방법은 고유 ID가 각각 부여된 문진 데이터를 생성하는 단계 -문진 데이터는 사용자 개인 정보 및 증상 및/또는 질환 관련 데이터 -; 사용자 단말로부터 문진 데이터에 대한 응답 데이터를 획득하는 단계; 및 전체 응답 데이터를 기 설정된 항목에 대응하는 고유 ID를 기초로 분류하고 기 설정된 항목에 대응하는 언어적 표현으로 추출해서 요약 데이터를 생성하는 단계;를 포함한다. An artificial intelligence-based questionnaire summary method executed by a processor of a server includes the steps of generating questionnaire data, each of which is assigned a unique ID - the questionnaire data is user personal information and symptom and/or disease-related data -; obtaining response data for the questionnaire data from a user terminal; and generating summary data by classifying the entire response data based on a unique ID corresponding to a preset item and extracting linguistic expressions corresponding to the preset item.","assignee":"주식회사 메디아크","inventors":["이찬형","이반형","이현성"],"publication_date":"2025-11-18","filing_date":"2025-11-10","priority_date":"2023-11-20","cpc_codes":["G","G16","G16H","G16H10/00","G16H10/20","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N20/00","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H15/00","G","G16","G16H","G16H50/00","G16H50/20"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250162477A/en"},{"publication_number":"KR20250162453A","title":"Apparatus and method for recommending outdoor advertising based on artificial intelligence","abstract":"본 발명은 옥외광고 추천 장치 및 방법에 관한 것이다. 상기 옥외광고 추천 장치는 유동인구에 대한 촬영 이미지, 소셜 미디어(Social Media)의 트렌드 정보 및 디스플레이 장치가 설치된 지역에 대한 지역 특성 정보를 획득하는 데이터 수집 모듈, 촬영 이미지를 분석하여, 유동인구에 관련된 파라미터(parameter)를 추출하는 추출 모듈 및 추출된 파라미터, 획득된 트렌드 정보 및 획득된 지역 특성 정보 중 적어도 하나와, 미리 학습된 딥러닝부를 기초로, 디스플레이 장치에 송출할 추천광고를 결정하는 결정 모듈을 포함하되, 트렌드 정보 및 지역 특성 정보는, 딥러닝부의 학습 과정 또는 딥러닝부의 출력값에 대한 후처리 과정에서 이용된다. The present invention relates to an outdoor advertisement recommendation device and method. The outdoor advertisement recommendation device comprises a data collection module that acquires a photographed image of a floating population, trend information of social media, and regional characteristic information of an area where a display device is installed, an extraction module that analyzes the photographed image to extract a parameter related to the floating population, and a decision module that determines a recommended advertisement to be transmitted to a display device based on at least one of the extracted parameter, the acquired trend information, and the acquired regional characteristic information, and a pre-learned deep learning unit, wherein the trend information and the regional characteristic information are used in a learning process of the deep learning unit or a post-processing process of an output value of the deep learning unit.","assignee":"주식회사 피치에이아이","inventors":["이동열","김서진"],"publication_date":"2025-11-18","filing_date":"2025-11-03","priority_date":"2022-08-02","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0242","G06Q30/0244","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q10/00","G06Q10/40","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0204","G06Q30/0205","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0242","G06Q30/0243","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0242","G06Q30/0246","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0247","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0241","G06Q30/0251","G06Q30/0268","G06Q50/01","G","G06","G06T","G06T7/00","G06T7/97","G","G06","G06V","G06V40/00","G06V40/10"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250162453A/en"},{"publication_number":"KR20250162446A","title":"Edge Computing-Based System and Method for Decentralized License Verification","abstract":"본 발명은 글로벌 네트워크 환경에서 중앙 집중식 인증의 긴 지연시간과 단일 실패 지점(SPOF) 문제를 해결하는 분산 엣지 기반 탈중앙화 라이선스 인증 시스템에 관한 것이다. 네트워크 라우팅 분산 장치, 글로벌 엣지 노드, 분산 데이터 저장소로 구성되며, 각 지역의 글로벌 엣지 노드에서 인증 요청을 독립적으로 처리하여 중앙 서버 의존성을 제거한다. 이를 통해 사용자 위치와 무관하게 저지연 인증을 제공하고, SPOF 가 없는 고가용성 구조로 안정성과 확장성을 제공한다. The present invention relates to a distributed edge-based decentralized license authentication system that addresses the long latency and single point of failure (SPOF) issues of centralized authentication in a global network environment. It comprises a network routing distribution device, a global edge node, and a distributed data storage system. Each global edge node independently processes authentication requests, eliminating reliance on a central server. This system provides low-latency authentication regardless of user location, while maintaining stability and scalability through a high-availability structure without a single point of failure.","assignee":"서준혁","inventors":["서준혁"],"publication_date":"2025-11-18","filing_date":"2025-11-01","priority_date":"2025-11-01","cpc_codes":["H","H04","H04L","H04L63/00","H04L63/08","H04L63/0884","G","G06","G06N","G06N20/00","H","H04","H04L","H04L63/00","H04L63/06","H04L63/067","H","H04","H04L","H04L63/00","H04L63/08","H04L63/0892","H","H04","H04L","H04L63/00","H04L63/16","H04L63/168","H","H04","H04L","H04L63/00","H04L63/20","H04L63/205","H","H04","H04L","H04L67/00","H04L67/01","H04L67/10","H04L67/1095"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250162446A/en"},{"publication_number":"KR20250161490A","title":"Method for detecting hazards and electronic device performing thereof","abstract":"차량에 대한 위험 감지를 위한 방법 및 전자 장치가 제공된다. 차량에 대한 위험 감지 방법은, 차량의 외부에 장착된 카메라를 이용하여 촬영된 원본 영상을 획득하는 동작, 원본 영상에 기초하여 딥러닝 영상 인식 방식에 대한 제1 처리 영상 및 컴퓨터비전 영상 인식 방식에 대한 제2 처리 영상을 각각 생성하는 동작, 딥러닝 영상 인식 모델을 이용하여 제1 처리 영상에 대한 딥러닝 인식 데이터를 생성하는 동작, 컴퓨터비전 영상 인식 모델을 이용하여 제2 처리 영상에 대해 컴퓨터비전 인식 데이터를 생성하는 동작, 및 딥러닝 인식 데이터 및 컴퓨터비전 인식 데이터에 기초하여 차량에 대한 위험을 감지하는 동작을 포함한다. A method and an electronic device for detecting a risk to a vehicle are provided. The method for detecting a risk to a vehicle includes: acquiring an original image captured using a camera mounted on the exterior of the vehicle; generating a first processed image for a deep learning image recognition method and a second processed image for a computer vision image recognition method based on the original image; generating deep learning recognition data for the first processed image using a deep learning image recognition model; generating computer vision recognition data for the second processed image using a computer vision image recognition model; and detecting a risk to the vehicle based on the deep learning recognition data and the computer vision recognition data.","assignee":"주식회사 앤씨앤","inventors":["라영찬","이용성"],"publication_date":"2025-11-17","filing_date":"2025-10-29","priority_date":"2023-02-15","cpc_codes":["B","B60","B60R","B60R11/00","B60R11/04","B","B60","B60W","B60W30/00","B60W30/08","B","B60","B60W","B60W40/00","B60W40/02","B","B60","B60W","B60W40/00","B60W40/10","B","B60","B60W","B60W50/00","B","B60","B60W","B60W50/00","B60W50/08","B60W50/14","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06V","G06V20/00","G06V20/50","G06V20/56","H","H04","H04N","H04N7/00","H04N7/18","B","B60","B60W","B60W2420/00","B60W2420/40","B60W2420/403","B","B60","B60W","B60W2554/00","B60W2554/20","B","B60","B60W","B60W2554/00","B60W2554/40"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250161490A/en"},{"publication_number":"KR20250161459A","title":"Device and method for analyzing importance based on safety incident prediction model","abstract":"본 개시의 실시예에 따른 인공지능을 기반으로 치매 노인의 안전 사고를 예측하는 모델을 생성하여 안전 사고를 예측하는 전자 장치는 메모리, 통신부, 및 상기 메모리 및 상기 통신부와 전기적으로 연결된 프로세서를 포함하고, 상기 프로세서는 상기 통신부를 통해 제1 사용자의 웨어러블 기기와 가정 내 설치된 모션 센서로부터 수집된 액티그래피 데이터 및 센싱 데이터를 수신하고, 오토 인코더를 통해 상기 액티그래피 데이터 및 센싱 데이터를 전처리한 데이터를 기반으로 상기 제1 사용자의 활동 패턴을 분류하며, 기설정된 학습 모델을 기반으로 상기 제1 사용자의 활동 패턴으로부터 상기 제1 사용자의 안전 사고를 예측하는 안전 사고 예측 모델을 추출하고, 상기 추출한 안전 사고 예측 모델을 기반으로 기설정된 주기에 따라 상기 제1 사용자의 안전 사고를 예측하도록 설정될 수 있다. An electronic device for predicting a safety accident by generating a model for predicting a safety accident of an elderly person with dementia based on artificial intelligence according to an embodiment of the present disclosure includes a memory, a communication unit, and a processor electrically connected to the memory and the communication unit, wherein the processor receives actigraphy data and sensing data collected from a wearable device of a first user and a motion sensor installed in a home through the communication unit, classifies an activity pattern of the first user based on data obtained by preprocessing the actigraphy data and sensing data through an autoencoder, extracts a safety accident prediction model for predicting a safety accident of the first user from the activity pattern of the first user based on a preset learning model, and predicts a safety accident of the first user according to a preset cycle based on the extracted safety accident prediction model.","assignee":"연세대학교 산학협력단; 가천대학교 산학협력단","inventors":["이경희","이지연","조애영","양은진"],"publication_date":"2025-11-17","filing_date":"2025-10-16","priority_date":"2023-10-24","cpc_codes":["G","G08","G08B","G08B31/00","A","A61","A61B","A61B5/00","A61B5/103","A61B5/11","A","A61","A61B","A61B5/00","A61B5/103","A61B5/11","A61B5/1116","A61B5/1117","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4806","A","A61","A61B","A61B5/00","A61B5/68","A61B5/6801","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G08","G08B","G08B21/00","G08B21/02","G08B21/04","G08B21/0407","G08B21/043","G","G08","G08B","G08B21/00","G08B21/02","G08B21/04","G08B21/0438","G08B21/0453","G","G08","G08B","G08B21/00","G08B21/02","G08B21/04","G08B21/0438","G08B21/0469","G","G08","G08B","G08B21/00","G08B21/02","G08B21/04","G08B21/0438","G08B21/0492","G","G08","G08B","G08B29/00","G08B29/18","G08B29/185","G","G16","G16H","G16H10/00","G16H10/20","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H50/00","G16H50/50","G","G16","G16H","G16H50/00","G16H50/70"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250161459A/en"},{"publication_number":"KR102884590B1","title":"Multi-model ai chat system and method thereof","abstract":"본 발명은 다중 모델 AI 채팅 시스템 및 그 방법에 관한 것으로서, 사용자가 입력한 사용자 메시지를 수신하는 사용자 입력 수신부, 기설정된 제1 인공지능 모델에 상기 사용자 메시지를 입력하여 초기 응답을 생성하는 제1 인공지능 모델부, 기설정된 제2 인공지능 모델에 상기 초기 응답을 입력하여 문법 오류, 비적절 표현, 언어 불일치 여부 검증 및 수정된 응답을 출력하는 제2 인공지능 모델부, 상기 사용자 메시지로부터 감정, 사실, 관계 및 맥락 정보를 추출하고 각 항목에 대응하는 중요도를 평가하는 사용자 메시지 평가부 및 상기 제2 인공지능 모델부(130)에서 생성한 수정 응답에 기설정된 상황 정보를 반영한 상황 반응형 응답을 생성하는 응답 생성부를 포함하는 것을 특징으로 한다. The present invention relates to a multi-model AI chat system and a method thereof, and is characterized by including a user input receiving unit for receiving a user message input by a user, a first artificial intelligence model unit for generating an initial response by inputting the user message into a first preset artificial intelligence model, a second artificial intelligence model unit for verifying the presence of grammatical errors, inappropriate expressions, and language inconsistencies by inputting the initial response into a second preset artificial intelligence model and outputting a modified response, a user message evaluation unit for extracting emotion, fact, relationship, and contextual information from the user message and evaluating the importance corresponding to each item, and a response generation unit for generating a situation-responsive response that reflects preset situational information in the modified response generated by the second artificial intelligence model unit (130).","assignee":"주식회사 블루노바랩","inventors":["배부순"],"publication_date":"2025-11-17","filing_date":"2025-06-04","priority_date":"2025-06-04","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/335","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F16/00","G06F16/30","G06F16/38","G","G06","G06F","G06F40/00","G06F40/20","G06F40/253","G","G06","G06F","G06F40/00","G06F40/30","G06F40/35","G","G06","G06N","G06N20/00"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102884590B1/en"},{"publication_number":"KR102887836B1","title":"Automatic Control Panel Manufacturing System","abstract":"본 발명은 자동 제어 판넬의 제조 기술에 관한 것으로, 보다 상세하게는 판넬 박스의 형성, 제어 부품의 설치, 각 부품 간의 전기적 배선, 센서 및 사용자 인터페이스 장치의 설치 등 자동 제어 판넬을 제조하는 일련의 공정을 자동화하고, 특히 인공지능 뉴럴 네트워크와 적응형 알고리즘을 적용하여 배선 공정의 정밀도, 품질 및 생산 효율성을 향상시키는 자동 제어 판넬 제조 시스템 및 그 방법에 관한 것이다. The present invention relates to a technology for manufacturing an automatic control panel, and more specifically, to an automatic control panel manufacturing system and method that automates a series of processes for manufacturing an automatic control panel, such as forming a panel box, installing control components, electrical wiring between components, and installing sensors and user interface devices, and in particular, improves the precision, quality, and production efficiency of the wiring process by applying an artificial intelligence neural network and an adaptive algorithm.","assignee":"김현우","inventors":["김현우"],"publication_date":"2025-11-17","filing_date":"2025-05-20","priority_date":"2025-05-20","cpc_codes":["G","G05","G05B","G05B19/00","G05B19/02","G05B19/418","G","G05","G05B","G05B19/00","G05B19/02","G05B19/18","G05B19/409","G","G06","G06F","G06F3/00","G06F3/01","G06F3/03","G06F3/041","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","H","H01","H01R","H01R43/00","H01R43/04","H01R43/048"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102887836B1/en"},{"publication_number":"LU601681B1","title":"Interactive Artificial Intelligence Voice Emotion Recognition System, Method, and Medium","abstract":"The present invention discloses an artificial intelligence voice emotion recognition system for interactive applications, comprising an interactive terminal and a cloud platform, wherein the interactive terminal is communicatively connected to the cloud platform. The interactive terminal integrates a multimodal data collection unit, an audio feature extraction unit, a multimodal emotion fusion unit, a user interaction interface, and a speech synthesizer. The cloud platform integrates an emotion feature database and an execution model, wherein the execution model outputs instruction information to the interactive terminal. Compared with the prior art, the present invention has the advantages of providing a highly intelligent system that utilizes multimodal data collection and analysis to ensure application accuracy, as well as a corresponding method and medium for the artificial intelligence voice emotion recognition system in interactive applications.","assignee":"Univ Xinyu","inventors":["Zhiping Zhang"],"publication_date":"2025-11-17","filing_date":"2025-05-17","priority_date":"2025-05-17","cpc_codes":["G","G10","G10L","G10L25/00","G10L25/48","G10L25/51","G10L25/63","A","A61","A61B","A61B5/00","A61B5/16","A61B5/165","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4803","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","G","G06","G06F","G06F3/00","G06F3/16","G06F3/167","G","G06","G06N","G06N3/00","G","G10","G10L","G10L25/00","G10L25/27","G10L25/30","A","A61","A61B","A61B2562/00","A61B2562/02","A61B2562/0204","A","A61","A61B","A61B5/00","A61B5/0059","A61B5/0077","A","A61","A61B","A61B5/00","A61B5/02","A61B5/0205","A","A61","A61B","A61B5/00","A61B5/02","A61B5/024","A61B5/02416","A","A61","A61B","A61B5/00","A61B5/05","A61B5/0507","A","A61","A61B","A61B5/00","A61B5/05","A61B5/053","A61B5/0531","A61B5/0533","A","A61","A61B","A61B5/00","A61B5/48","A61B5/486","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7275"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU601681B1/en"},{"publication_number":"LU601665B1","title":"Data-Augmented Contrastive Learning-Based Anomaly Diagnosis Method and System for Cloud Servers","abstract":"The present invention discloses a data-augmented contrastive learning-based anomaly diagnosis method, system for cloud servers. The anomaly diagnosis method includes: S1: performing standardization processing on input server performance time- series data; S2: designing a dual-attention view model based on data-augmented contrastive learning using data augmentation, generating a new branch from attention representations of cloud server performance time-series data, distinguishing normal points from abnormal points by measuring the distance between two attention branches; S3: designing an asymmetric KL divergence loss function based on the similarity of the two branches in the dual-attention view structure; S4: training the designed model using historical data; S5: applying the trained model to real-time server performance data to determine whether data points are abnormal, and obtaining detailed anomaly reports for detected abnormal points; S6: periodically updating and optimizing the model based on anomaly reports and feedback from operation personnel.","assignee":"Univ Taiyuan Science & Tech","inventors":["Yinzhang Guo"],"publication_date":"2025-11-17","filing_date":"2025-05-16","priority_date":"2025-05-16","cpc_codes":["H","H04","H04L","H04L41/00","H04L41/14","H04L41/142","G","G06","G06F","G06F11/00","G06F11/30","G","G06","G06N","G06N3/00","H","H04","H04L","H04L41/00","H04L41/16","H","H04","H04L","H04L43/00","H04L43/08","H04L43/0805","H04L43/0817"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU601665B1/en"},{"publication_number":"KR20250161434A","title":"Exon-junction markers and reference markers for cancer diagnosis","abstract":"본 발명은 엑손-접합부(exon-junction)를 포함하는 암 진단용 마커 및 레퍼런스 마커에 관한 것으로, 보다 상세하게는 혈액 내 RNA의 엑손-접합부를 포함하는 암 진단용 마커 및 레퍼런스 마커에 관한 것이다. 엑손-접합부를 포함하는 본 발명의 암 진단 마커 및 이를 정규화하기 위한 레퍼런스 마커를 활용하면 암 특이적인 스플라이싱 사건을 더 잘 탐지하고, 오염된 유전체 DNA의 간섭을 줄일 수 있고, 낮은 농도의 전사체를 보다 정확하게 정량화하여 종양 특이적인 RNA 시그니쳐를 더 잘 구별할 수 있다. 또한, 초기 단계 질병과 관련된 미세한 분자 변화를 탐지하는 민감도를 높여, 초기 암 진단을 위한 강력한 도구가 될 수 있다. The present invention relates to a cancer diagnostic marker and reference marker including an exon-junction, and more particularly, to a cancer diagnostic marker and reference marker including an exon-junction of RNA in blood. Utilizing the cancer diagnostic markers of the present invention, including exon junctions, and reference markers for normalizing them, can better detect cancer-specific splicing events, reduce interference from contaminating genomic DNA, and more accurately quantify low-abundance transcripts, thereby better distinguishing tumor-specific RNA signatures. Furthermore, by increasing the sensitivity for detecting subtle molecular changes associated with early-stage disease, the markers can become powerful tools for early-stage cancer diagnosis.","assignee":"주식회사 포어텔마이헬스","inventors":["안태진","안은용","박성민","김사라","김현정","이혜진"],"publication_date":"2025-11-17","filing_date":"2025-05-07","priority_date":"2024-05-07","cpc_codes":["C","C12","C12Q","C12Q1/00","C12Q1/68","C12Q1/6876","C12Q1/6883","C12Q1/6886","C","C12","C12Q","C12Q1/00","C12Q1/68","C12Q1/6844","C12Q1/686","C","C12","C12Q","C12Q1/00","C12Q1/68","C12Q1/6869","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G16","G16B","G16B25/00","G16B25/10","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/70","C","C12","C12Q","C12Q2531/00","C12Q2531/10","C12Q2531/113","C","C12","C12Q","C12Q2537/00","C12Q2537/10","C12Q2537/165","C","C12","C12Q","C12Q2600/00","C12Q2600/158"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250161434A/en"},{"publication_number":"KR102887148B1","title":"Method and apparatus for verification of fake video through artificial intelligence model","abstract":"본 개시는 전자 장치가 딥페이크 영상을 검증하는 방법 및 이를 수행하는 전자 장치에 관한 것이다. 일 실시 예에 의하면, 전자 장치가 딥페이크 영상을 검증하는 방법은 검증 대상 영상을 획득하는 단계; 상기 획득된 검증 대상 영상 내 소정의 프레임 간격을 가지는 프레임 이미지들을 획득하는 단계; 상기 획득된 프레임 이미지들이 입력되면, 상기 프레임 이미지들에 대한 특징 벡터를 출력하는 제1 인공 지능 모델에 상기 프레임 이미지들을 입력함으로써, 상기 제1 인공 지능 모델로부터 상기 프레임 이미지들 각각에 대한 특징 벡터들을 획득하는 단계; 및 상기 획득된 특징 벡터들이 입력되면 상기 검증 대상 영상이 딥페이크 영상인지 여부에 대한 결과 정보를 출력하는 제2 인공 지능 모델에, 상기 특징 벡터들을 입력함으로써 상기 제2 인공 지능 모델로부터 상기 결과 정보를 획득하는 단계; 를 포함할 수 있다. The present disclosure relates to a method for an electronic device to verify a deepfake image and an electronic device performing the same. According to one embodiment, the method for an electronic device to verify a deepfake image may include the steps of: acquiring a verification target image; acquiring frame images having a predetermined frame interval within the acquired verification target image; inputting the acquired frame images into a first artificial intelligence model that outputs feature vectors for the frame images when the acquired frame images are input, thereby acquiring feature vectors for each of the frame images from the first artificial intelligence model; and inputting the feature vectors into a second artificial intelligence model that outputs result information on whether the verification target image is a deepfake image when the acquired feature vectors are input, thereby acquiring result information from the second artificial intelligence model.","assignee":"울산과학기술원","inventors":["백승렬","김동욱"],"publication_date":"2025-11-17","filing_date":"2022-01-21","priority_date":"2022-01-21","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06T","G06T7/00","G06T7/0002","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V40/00","G06V40/40","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102887148B1/en"},{"publication_number":"KR102886396B1","title":"Apparatus and method for detecting abnormal fixed bolt","abstract":"대상물을 고정하는 고정 볼트의 이상을 탐지하는 인공지능 기반 고정 볼트 이상 탐지 장치 및 방법에 관한 것으로, 메모리, 그리고 하나 이상의 코어를 포함하는 프로세서를 포함하고, 상기 프로세서는, 고정 볼트를 포함하는 대상물 이미지를 획득하고, 상기 획득한 대상물 이미지로부터 상기 고정 볼트 이상 탐지를 위한 후보 지점을 추출하며, 상기 대상물 이미지로부터 추출한 후보 지점을 포함하는 관심 영역을 컷팅(cutting)하고, 상기 컷팅한 관심 영역을 사전 학습된 뉴럴 네트워크 모델에 입력하여 상기 후보 지점에 대한 고정 볼트 이상 유무를 판별하며, 상기 고정 볼트 이상으로 판별된 후보 지점에 대한 결과 정보를 생성하는 것을 특징으로 한다. The present invention relates to an artificial intelligence-based fixed bolt abnormality detection device and method for detecting an abnormality in a fixed bolt fixing an object, comprising: a memory; and a processor including one or more cores, wherein the processor acquires an object image including a fixed bolt, extracts a candidate point for detecting the fixed bolt abnormality from the acquired object image, cuts a region of interest including the candidate point extracted from the object image, inputs the cut region of interest into a pre-trained neural network model to determine whether there is an abnormality in the fixed bolt for the candidate point, and generates result information for the candidate point determined to be an abnormality in the fixed bolt.","assignee":"(주)이포즌","inventors":["김영진","손정모","김수민"],"publication_date":"2025-11-17","filing_date":"2022-07-12","priority_date":"2022-07-12","cpc_codes":["G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","B","B64","B64C","B64C39/00","B64C39/02","B64C39/024","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","B","B64","B64U","B64U2101/00","B64U2101/30","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G01N2021/8854","G01N2021/8861","G01N2021/8864","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G01N2021/8883","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G01N2021/8887","G","G01","G01N","G01N2201/00","G01N2201/12","G01N2201/129","G01N2201/1296"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102886396B1/en"},{"publication_number":"KR102887583B1","title":"Artificial intelligence system for reflecting skill level of user and operation method thereof","abstract":"본 개시는 서버, 관리자 단말기 및 사용자 단말기를 포함하는 사용자의 숙련도를 반영하기 위한 인공지능 시스템에 관한 것으로써, 인공지능 시스템의 동작 방법은, 서버가 입력 영상을 획득하는 단계, 서버가 영상으로부터 미리 정해진 물체를 감지하기 위한 기계학습모델을 획득하는 단계, 서버가 입력 영상을 기계학습모델에 적용하여 입력 영상에서 미리 정해진 물체의 위치 및 크기와 관련된 출력 정보를 획득하는 단계, 사용자 단말기가 입력 영상, 출력 정보 및 사용자 인터페이스를 표시하는 단계, 및 사용자 단말기가 사용자로부터 출력 정보에 대한 피드백 정보를 수신하는 단계를 포함하고, 피드백 정보는 출력 정보가 옳은지 여부를 나타낸다. The present disclosure relates to an artificial intelligence system for reflecting a user's proficiency, including a server, an administrator terminal, and a user terminal, and a method for operating the artificial intelligence system, including a step in which a server obtains an input image, a step in which the server obtains a machine learning model for detecting a predetermined object from the image, a step in which the server applies the input image to the machine learning model to obtain output information related to the position and size of the predetermined object from the input image, a step in which the user terminal displays the input image, the output information, and a user interface, and a step in which the user terminal receives feedback information on the output information from the user, wherein the feedback information indicates whether the output information is correct.","assignee":"광운대학교 산학협력단","inventors":["김현경","박선영"],"publication_date":"2025-11-17","filing_date":"2021-12-07","priority_date":"2021-12-07","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06F","G06F18/00","G06F18/40","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T7/00","G06T7/60","G06T7/62","G","G06","G06T","G06T7/00","G06T7/70","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102887583B1/en"},{"publication_number":"KR102885487B1","title":"Art asset valuation methods and systems using artificial intelligence","abstract":"인공지능을 이용한 미술 자산 가치 평가 시스템은 인스트럭션들을 저장하는 메모리 및 프로세서를 포함할 수 있다. 인스트럭션들은, 상기 프로세서에 의해 실행 시에, 상기 시스템이 복수의 데이터 소스로부터 작가 정보, 작품 특성 정보, 및 시장 거래 정보를 포함하는 미술 작품 관련 데이터를 수집하고, 수집된 데이터를 정규화하여 학습 데이터를 생성하며, 학습 데이터를 이용하여 현재 가치 예측 모델, 가격 변동 예측 모델, 및 투자 가치 평가 모델을 학습하고, 학습된 모델들을 이용하여 특정 미술 작품의 현재 시장 가치, 미래 가격 변동 예측 및 장기 투자 가치를 산출하며, 산출된 결과들을 통합하여 최종 평가 결과를 생성하도록 제어할 수 있다. An art asset valuation system using artificial intelligence may include a memory and a processor storing instructions. The instructions, when executed by the processor, may control the system to collect art-related data, including artist information, artwork characteristics, and market transaction information, from multiple data sources, normalize the collected data to generate training data, use the training data to train a current value prediction model, a price fluctuation prediction model, and an investment value assessment model, use the trained models to calculate the current market value, future price fluctuation prediction, and long-term investment value of a specific art piece, and integrate the calculated results to generate a final valuation result.","assignee":"주식회사 에버트레져","inventors":["조영린","임원진","박산하"],"publication_date":"2025-11-17","filing_date":"2024-11-28","priority_date":"2024-11-28","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0278","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0202","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0203","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0206","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0283","G","G06","G06Q","G06Q40/00","G06Q40/04","G","G06","G06Q","G06Q50/00","G06Q50/10"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102885487B1/en"},{"publication_number":"ES3041985T3","title":"Data processing system for generating predictions of cognitive outcome in patients","abstract":"Un sistema para generar una representación visual del cerebro de un paciente está configurado para recibir datos de sensores que representan el comportamiento de una región cerebral. El sistema recupera datos de mapeo que asocian un valor de predicción a dicha región. Este valor de predicción indica el efecto que tendrá un tratamiento en la región sobre el comportamiento del paciente, y los datos de mapeo están indexados a un identificador del paciente. Tras la aplicación de un estímulo a la región, el sistema recibe datos de sensores que representan su comportamiento. El sistema ejecuta un modelo que actualiza, basándose en los datos de los sensores, el valor de predicción de la región. Al ejecutar el modelo, el sistema actualiza los datos de mapeo incluyendo el valor de predicción actualizado. Finalmente, el sistema genera una representación visual de los datos de mapeo actualizados, que incluye el valor de predicción actualizado. (Traducción automática con Google Translate, sin valor legal) A system for generating a visual representation of a patient's brain is configured to receive sensor data that represents the behavior of a brain region. The system retrieves mapping data that associates a prediction value with that region. This prediction value indicates the effect that a treatment in the region will have on the patient's behavior, and the mapping data is indexed to a patient identifier. After a stimulus is applied to the region, the system receives sensor data that represents its behavior. The system runs a model that updates the region's prediction value based on the sensor data. When the model is run, the system updates the mapping data to include the updated prediction value. Finally, the system generates a visual representation of the updated mapping data, including the updated prediction value.","assignee":"Carnegie Mellon University","inventors":["Bradford Mahon","Keith Parkins","Max Sims","Benjamin Chernoff","Hugo Angulo-Orquera"],"publication_date":"2025-11-17","filing_date":"2019-12-02","priority_date":"2018-11-30","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/70","A","A61","A61N","A61N1/00","A61N1/02","A61N1/04","A61N1/05","A61N1/0526","A61N1/0529","A","A61","A61B","A61B34/00","A61B34/10","A","A61","A61B","A61B5/00","A61B5/0033","A61B5/004","A61B5/0042","A","A61","A61B","A61B5/00","A61B5/0059","A61B5/0077","A","A61","A61B","A61B5/00","A61B5/05","A61B5/055","A","A61","A61B","A61B5/00","A61B5/24","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/369","A61B5/372","A","A61","A61B","A61B5/00","A61B5/40","A61B5/4058","A61B5/4064","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4848","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","A","A61","A61B","A61B5/00","A61B5/74","A61B5/742","A61B5/7425","A","A61","A61N","A61N1/00","A61N1/18","A61N1/32","A61N1/36","A61N1/36014","A","A61","A61N","A61N1/00","A61N1/18","A61N1/32","A61N1/36","A61N1/362","A61N1/37","G","G01","G01R","G01R33/00","G01R33/20","G01R33/44","G01R33/48","G01R33/4806","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H20/00","G16H20/30","G"],"country":"ES","kind":"application","source_url":"https://patents.google.com/patent/ES3041985T3/en"},{"publication_number":"ES3041900T3","title":"Architecture for block sparse operations on a systolic array","abstract":"Las implementaciones descritas aquí incluyen software, firmware y lógica de hardware que proporcionan técnicas para realizar operaciones aritméticas con datos dispersos mediante una unidad de procesamiento sistólico. Una implementación proporciona información sobre la dispersión de los datos mediante flujos de bits comprimidos. Otra implementación proporciona instrucciones de producto escalar para datos dispersos en bloque. Una tercera implementación proporciona un adaptador de profundidad para una matriz sistólica. (Traducción automática con Google Translate, sin valor legal) The implementations described here include software, firmware, and hardware logic that provide techniques for performing arithmetic operations on sparse data using a systolic processing unit. One implementation provides information about data sparseness using compressed bitstreams. Another implementation provides dot product instructions for block sparse data. A third implementation provides a depth adapter for a systolic array.","assignee":"Intel Corp","inventors":["Abhishek Appu","Subramaniam Maiyuran","Mike Macpherson","Fangwen Fu","Jiasheng Chen","Varghese George","Vasanth Ranganathan","Ashutosh Garg","Joydeep Ray"],"publication_date":"2025-11-17","filing_date":"2020-03-14","priority_date":"2019-03-15","cpc_codes":["G","G06","G06F","G06F12/00","G06F12/02","G06F12/08","G06F12/0802","G06F12/0806","G","G06","G06F","G06F15/00","G06F15/76","G06F15/80","G06F15/8007","G","G06","G06F","G06F15/00","G06F15/76","G06F15/80","G06F15/8046","G","G06","G06F","G06F17/00","G06F17/10","G06F17/16","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/544","G06F7/5443","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/30003","G06F9/30007","G06F9/3001","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/30003","G06F9/30007","G06F9/30036","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/30145","G06F9/3016","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/38","G06F9/3836","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/38","G06F9/3885","G06F9/3887","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/38","G06F9/3885","G06F9/3888","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/38","G06F9/3885","G06F9/3888","G06F9/38885","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5027","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02"],"country":"ES","kind":"application","source_url":"https://patents.google.com/patent/ES3041900T3/en"},{"publication_number":"TW202544693A","title":"Dynamically determining optimal selling price of unsold performance tickets server and operation method thereof","abstract":"Provided are a server and an operation method for dynamically determining an optimal selling price of unsold performance tickets. The server comprises a memory, a communication module, and a processor. The processor is configured to collect first ticket-related data and first performance-related data of performance tickets for a plurality of performance seats at least at one point in time, input the first ticket-related data and the first performance-related data into a first artificial intelligence model to predict expected sales data of the performance tickets at the time of closing sales of the performance tickets, and when the expected sales data is lower than a predetermined threshold, input the remaining time until the time of closing sales for unsold performance tickets, the remaining quantity, second ticket-related data, and second performance-related data into a second artificial intelligence model to determine an optimal selling price of the unsold performance tickets.","assignee":"南韓商諾爾宇宙股份有限公司","inventors":["鄭光炯"],"publication_date":"2025-11-16","filing_date":"2025-05-12","priority_date":"2024-05-10","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0206","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0202","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0283"],"country":"TW","kind":"application","source_url":"https://patents.google.com/patent/TW202544693A/en"},{"publication_number":"TW202544686A","title":"Mixed analog/digital in-memory computing system with noise reduction and the method of the same","abstract":"A mixed analog/digital in-memory computing device implements matrix vector multiplication with reduced noise for use by a deep neural network (DNN). For each row of a cross-bar array a multiplier is split into at least a most significant (MS) portion and a least significant (LS) portion and preloaded into at least two cells on one row and at least two different columns of the cross-bar array. An input activation (IA) value is driven onto input conductors of each row and an analog-to-digital converter (ADC) converts output signals from the two columns as a truncated MS partial sum and a truncated LS partial sum. A gain is applied to the truncated MS partial sum and added to the truncated LS partial sum to form a resulting value for one node of the DNN.","assignee":"美商豪威科技股份有限公司","inventors":["齊藤大輔"],"publication_date":"2025-11-16","filing_date":"2025-05-02","priority_date":"2024-05-03","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G06N3/065","G","G06","G06F","G06F17/00","G06F17/10","G06F17/16","G","G06","G06F","G06F3/00","G06F3/06","G06F3/0601","G06F3/0602","G06F3/0625","G","G06","G06F","G06F3/00","G06F3/06","G06F3/0601","G06F3/0628","G06F3/0655","G06F3/0659","G","G06","G06F","G06F3/00","G06F3/06","G06F3/0601","G06F3/0668","G06F3/0671","G06F3/0673","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/52","G06F7/523","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/544","G06F7/5443","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/57","G06F7/575","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V10/00","G06V10/94"],"country":"TW","kind":"application","source_url":"https://patents.google.com/patent/TW202544686A/en"},{"publication_number":"TW202544606A","title":"Adaptive power management for ai/ml accelerators","abstract":"Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing dynamic voltage and/or frequency scaling. One of the methods includes obtaining an input data item; processing the input data item using a data processing pipeline executed on the computing device to generate an output data item, wherein processing the input data item comprises: executing a sequence of one or more pre-processing operations of the data processing pipeline on the input data item to generate a pre-processed data item; during the executing, monitoring a length of time consumed in executing the one or more pre-processing operations; and determining, based on the length of time consumed in executing the one or more pre-processing operations, a voltage value for a target processing core that will execute one or more subsequent processing operations of the data processing pipeline to process the pre-processed data item to generate a processed data item.","assignee":"美商谷歌有限責任公司","inventors":["李康民","朴熙俊","賈加迪什 巴斯卡 帕卡拉沃爾"],"publication_date":"2025-11-16","filing_date":"2025-04-11","priority_date":"2024-04-11","cpc_codes":["G","G06","G06F","G06F1/00","G06F1/26","G06F1/32","G06F1/3203","G06F1/3234","G06F1/3296","G","G06","G06F","G06F1/00","G06F1/26","G06F1/32","G06F1/3203","G06F1/3206","G06F1/3228","G","G06","G06F","G06F1/00","G06F1/26","G06F1/32","G06F1/3203","G06F1/3234","G06F1/324","G","G06","G06F","G06F1/00","G06F1/26","G06F1/32","G06F1/3203","G06F1/3234","G06F1/3243","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/48","G06F9/4806","G06F9/4843","G06F9/4881","G06F9/4893","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5094","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/501","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/508","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/509","Y","Y02","Y02D","Y02D10/00"],"country":"TW","kind":"application","source_url":"https://patents.google.com/patent/TW202544606A/en"},{"publication_number":"TW202544685A","title":"Sparse high rank adapters and their hardware-software co-design","abstract":"A processor-implemented method includes receiving an artificial neural network having a number of pre-trained weights. The method also includes training a subset of the number of pre-trained weights to obtain trained sparse adapter weights for obtaining a fine-tuned version of the artificial neural network. The subset of the number of pre-trained weights includes base model weights of a base model for the artificial neural network. The subset of the number of pre-trained weights is selected with a sparse mask of a sparse adapter. The method may also include replacing the subset of the number of pre-trained weights with the trained sparse adapter weights.","assignee":"美商高通公司","inventors":["卡提凱亞 巴德瓦傑","尼爾什普拉薩德 潘迪","斯韋塔 普里亞達什","舒布漢卡爾曼格甚 博爾斯","什雷亞 卡丹比","拉斐爾澤維爾 埃斯蒂夫","維斯瓦納特 加納帕蒂","里希克 加瑞帕立","保羅尼可拉斯 霍特莫夫","馬里納斯威廉 范巴倫","馬庫斯 納格爾"],"publication_date":"2025-11-16","filing_date":"2025-03-14","priority_date":"2024-05-14","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096"],"country":"TW","kind":"application","source_url":"https://patents.google.com/patent/TW202544685A/en"},{"publication_number":"TW202544674A","title":"Artificial intelligence system and method for transistor-level place and route of flattened digital circuits","abstract":"In an integrated circuit, a method of flattened placement and/or routing of multiple transistors using an artificial intelligence (AI) model, comprising the steps of receiving an input consisting of a circuit schematic and features; performing supervised or unsupervised learning of the AI model, to produce a predicted placement and/or routing by the AI model; using a cost function for optimizing transistor placement and/or routing. An optimal placement and/or routing is generated by iteratively running cycles of placement and/or routing, based on the AI model; assigning a value to each placement and/or routing combination, according to the cost function and recording the placements and/or routings and value information across the performed iterations until meeting a predetermined metric or exceeding a predetermined number of iterations or elapsed time; ranking each of the generated placements and/or routing according to the cost function; selecting the k-top (k=1,2,….) placement and/or routing combinations.","assignee":"以色列商尼歐邏輯有限公司","inventors":["阿維 梅西卡","齊夫 萊謝姆","阿里耶 福斯特","阿隆 特特羅"],"publication_date":"2025-11-16","filing_date":"2025-04-02","priority_date":"2024-04-03","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06F","G06F30/00","G06F30/30","G06F30/39","G06F30/392","G","G06","G06F","G06F30/00","G06F30/30","G06F30/39","G06F30/398","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/12","G06N3/126","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06F","G06F2111/00","G06F2111/08"],"country":"TW","kind":"application","source_url":"https://patents.google.com/patent/TW202544674A/en"},{"publication_number":"KR20250160861A","title":"Method for diagnosising cardiovascular disease and device using the same","abstract":"심혈관 질환 진단 방법 및 이를 이용한 장치가 개시된다. 일 실시예에 따른 진단 장치의 제어방법은 피검체의 망막 이미지를 획득하는 단계; 및 상기 망막 이미지를 기초로 머신러닝 모델을 이용하여 상기 대상체에 대한 심혈관 질환 진단 정보를 획득하는 단계를 포함하고, 상기 머신러닝 모델은 제1 모델 및 제2 모델을 포함하고, 상기 제1 모델은 신경망 모델이고, 상기 제2 모델은 회귀 기반의 머신러닝 모델일 수 있다. A method for diagnosing cardiovascular disease and a device using the same are disclosed. A method for controlling a diagnostic device according to one embodiment includes the steps of: obtaining a retinal image of a subject; and obtaining cardiovascular disease diagnosis information for the subject using a machine learning model based on the retinal image. The machine learning model may include a first model and a second model, wherein the first model may be a neural network model and the second model may be a regression-based machine learning model.","assignee":"주식회사 메디웨일","inventors":["최태근","이근영","임형택"],"publication_date":"2025-11-14","filing_date":"2025-10-27","priority_date":"2023-10-29","cpc_codes":["A","A61","A61B","A61B3/00","A61B3/0016","A61B3/0025","A","A61","A61B","A61B3/00","A","A61","A61B","A61B3/00","A61B3/10","A61B3/12","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06V","G06V10/00","G06V10/70","G06V10/766","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V40/00","G06V40/10","G06V40/18","G","G16","G16H","G16H20/00","G","G16","G16H","G16H30/00","G16H30/20","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/70","G","G16","G16H","G16H70/00","G16H70/20"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250160861A/en"},{"publication_number":"CN120954241A","title":"Intelligent Traffic Early Warning System and Method Based on Video Inspection and Flashing Warning Linkage","abstract":"The invention discloses an intelligent traffic early warning system and method based on video inspection and explosion flash warning linkage, and relates to the technical field of intelligent traffic systems, wherein the system comprises a video inspection module, an explosion flash warning module, an edge computing unit and a cloud platform; the video inspection module is used for monitoring traffic conditions within a set range of a one-way lane and comprises a zoom dome camera unit and a fixed Jiao Qiangji unit which is symmetrically arranged front and back, the flash warning module comprises a conventional and multifunctional device, the edge computing unit processes video data through an improved visual model, and the cloud platform executes event classification and space matching to generate a linkage control instruction. The invention has the advantages of accurately detecting abnormal events in real time, dynamically adjusting early warning parameters and realizing multi-dimensional cooperative linkage.","assignee":"Hebei Jiaotou Intelligent Technology Co ltd","inventors":["李广","张逸舟","杨帆","付国龙","田森","陈伟","孙计山","尹春辉","黄伟亮","袁瑞"],"publication_date":"2025-11-14","filing_date":"2025-10-20","priority_date":"2025-10-20","cpc_codes":["G","G08","G08G","G08G1/00","G08G1/01","G08G1/0104","G08G1/0108","G08G1/0116","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/40","G06V20/41","G","G06","G06V","G06V20/00","G06V20/40","G06V20/44","G","G06","G06V","G06V20/00","G06V20/50","G06V20/52","G06V20/54","G","G08","G08B","G08B21/00","G08B21/18","G08B21/182","G","G08","G08B","G08B29/00","G08B29/18","G08B29/185","G08B29/188","G","G08","G08B","G08B31/00","G","G08","G08B","G08B7/00","G08B7/06","G","G08","G08G","G08G1/00","G08G1/01","G08G1/0104","G08G1/0125","G08G1/0133","G","G08","G08G","G08G1/00","G08G1/01","G08G1/0104","G08G1/0137","G08G1/0145"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120954241A/en"},{"publication_number":"CN120954208A","title":"A Beidou video multi-behavior analysis and early warning system and terminal","abstract":"本发明涉及视频监控技术领域，公开了一种北斗视频多行为分析预警系统及终端，视频采集模块，用于构建时空同步的视频监控框架，采用自适应图像增强算法对采集的视频流进行预处理，得到预处理后视频；多模态特征融合模块，用于利用3D‑CNN网络从预处理后视频中提取行人局部时空特征，并结合HRNet模型获取人体特征坐标，将北斗定位的摄像头坐标与人体特征坐标映射到统一地理坐标系，得到多模态特征集；行为识别模块，用于基于ST‑GCN网络建模行人肢体的动作序列，结合Transformer编码器捕捉长期依赖关系，确定为异常行为及对应等级；预警模块，用于根据异常行为等级触发对应的预警机制，并推送异常信息；本发明提升预警效率。 This invention relates to the field of video surveillance technology and discloses a BeiDou video multi-behavior analysis and early warning system and terminal. The system includes a video acquisition module for constructing a spatiotemporally synchronized video surveillance framework, employing an adaptive image enhancement algorithm to preprocess the acquired video stream to obtain preprocessed video; a multimodal feature fusion module for extracting local spatiotemporal features of pedestrians from the preprocessed video using a 3D-CNN network, and combining this with an HRNet model to obtain human body feature coordinates, mapping the BeiDou-positioned camera coordinates and human body feature coordinates to a unified geographic coordinate system to obtain a multimodal feature set; a behavior recognition module for modeling pedestrian limb movement sequences based on an ST-GCN network, combining this with a Transformer encoder to capture long-term dependencies, and identifying abnormal behaviors and their corresponding levels; and an early warning module for triggering corresponding early warning mechanisms based on the abnormal behavior level and pushing abnormal information. This invention improves early warning efficiency.","assignee":"Jiangsu second normal university; Anhui Zhengxin Information Technology Co ltd; Anhui University","inventors":["陈思宝","倪艺洋","张旭波","江淮","李骏","吴屹铭","高道畅"],"publication_date":"2025-11-14","filing_date":"2025-10-20","priority_date":"2025-10-20","cpc_codes":["G","G08","G08B","G08B31/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/40","G06V10/62","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/7715","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/40","G06V20/44","G","G06","G06V","G06V20/00","G06V20/40","G06V20/46","G","G06","G06V","G06V20/00","G06V20/50","G06V20/52","G","G06","G06V","G06V40/00","G06V40/20","G","G08","G08B","G08B25/00","G08B25/01","G08B25/10","G","G08","G08B","G08B29/00","G08B29/18","G08B29/185","G08B29/186","G","G08","G08B","G08B29/00","G08B29/18","G08B29/185","G08B29/188"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120954208A/en"},{"publication_number":"CN120954500A","title":"Spatial Domain Identification Method for Spatial Transcriptome Data Using Multi-Spatial Self-Supervised Contrast Learning","abstract":"本发明公开了多空间自监督对比学习的空间转录组数据空间域识别方法，包括：建模生成空间邻域图，保持图结构不变，打乱节点特征进行数据增强，作为增强图；基于图神经网络构建编码器，对融合了空间信息与基因信息的空间转录组数据进行提取得到潜在嵌入，送入多空间生成器以生成多组丰富的图特征表示；将图特征表示与潜在嵌入融合得到精炼的表示，通过解码器重构成基因表达矩阵，将对比学习损失和重建损失两者加和作为总目标函数；根据上述所得总目标函数，采用Adam优化器对网络参数进行更新，完成空间转录组空间域识别。本发明从全局和局部角度充分挖掘空间转录组数据，实现精确的空间域识别。 This invention discloses a spatial domain identification method for spatial transcriptome data using multi-spatial self-supervised contrastive learning, comprising: modeling and generating a spatial neighborhood graph, maintaining the graph structure, shuffling node features for data augmentation, and using this as an augmented graph; constructing an encoder based on a graph neural network to extract latent embeddings from spatial transcriptome data that integrates spatial and genetic information, and feeding these latent embeddings into a multi-spatial generator to generate multiple sets of rich graph feature representations; fusing the graph feature representations with the latent embeddings to obtain a refined representation, reconstructing the gene expression matrix through a decoder, and using the sum of the contrastive learning loss and the reconstruction loss as the overall objective function; and updating the network parameters using the Adam optimizer based on the obtained overall objective function to complete the spatial domain identification of the spatial transcriptome. This invention fully mines spatial transcriptome data from both global and local perspectives to achieve accurate spatial domain identification.","assignee":"Anhui University","inventors":["丁云","何锐","侯明扬","钟嘉成","戴殷强","郑春厚"],"publication_date":"2025-11-14","filing_date":"2025-10-20","priority_date":"2025-10-20","cpc_codes":["G","G16","G16B","G16B25/00","G16B25/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G16","G16B","G16B30/00","G","G16","G16B","G16B40/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120954500A/en"},{"publication_number":"CN120950929A","title":"Methods and Systems for Constructing a Network Traffic Feature Information Database for Intelligent Power Devices","abstract":"The embodiment of the invention provides a method and a system for constructing a network flow characteristic information base of electric intelligent equipment, belonging to the technical field of electric digital signal processing. The construction method comprises the steps of obtaining network flow information to be processed, constructing a space-time diagram according to the network flow information, calculating similarity of each node in the space-time diagram by adopting a dynamic time regulation method to determine the weight of edges between every two nodes, extracting space-time dependent feature vectors in the space-time diagram by adopting a space-time attention network, training and learning the space-time dependent feature vectors by adopting a federal self-supervision feature learning method to obtain corresponding robust feature vectors, determining feature subsets by adopting a dynamic feature selector according to the robust feature vectors, and constructing a feature information base based on the feature subsets. The construction method and the construction system can construct the network flow characteristic information base of the power equipment, which is suitable for diversified network flows.","assignee":"State Grid Siji Network Security Beijing Co ltd; State Grid Siji Testing Technology Beijing Co Ltd","inventors":["李永刚","王利斌","潘善民","韩淞","任磊","佟雪松","刘继涛","刘一霖","谷五勋","李鑫","马建勋","刘霞飞"],"publication_date":"2025-11-14","filing_date":"2025-10-20","priority_date":"2025-10-20","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/211","G06F18/2113","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2137","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120950929A/en"},{"publication_number":"CN120950908A","title":"A Deep Learning-Based Method and System for Predicting the Lifespan of Kitchen Air Conditioner Filter Devices","abstract":"The invention relates to the technical field of life prediction of a kitchen air conditioner filter device, in particular to a life prediction method and a system of the kitchen air conditioner filter device based on deep learning, wherein the method comprises the steps that edge equipment collects real-time kitchen air conditioner operation parameters and extracts basic physical characteristic indexes of the filter device; the edge gateway deploys a first local PINN-transducer model, performs data preprocessing and local life prediction reasoning, caches incremental gradient data, and uploads the incremental gradient data to the cloud based on a safety fault tolerance mechanism, deploys a second local PINN-transducer model, performs distributed training by adopting a hybrid parallel architecture, synchronously updates the global PINN-transducer model, and sends model parameter packets to each edge node after signature encryption. By constructing a filter screen life prediction system integrating a physical mechanism and time sequence learning, the prediction precision and generalization capability of a high-fluctuation oil smoke scene can be improved.","assignee":"Cheari Beijing Certification & Testing Co ltd; Qingdao Haier Air Conditioner Gen Corp Ltd","inventors":["王超","李欣","蔡宁","高孺","王伯燕","张子祺","杨双","张宇佳","王志坤"],"publication_date":"2025-11-14","filing_date":"2025-10-20","priority_date":"2025-10-20","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","F","F24","F24F","F24F11/00","F24F11/30","F24F11/32","F24F11/39","F","F24","F24F","F24F11/00","F24F11/50","F24F11/56","F24F11/58","F","F24","F24F","F24F11/00","F24F11/62","F24F11/63","F24F11/64","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N5/00","G06N5/04","G06N5/043","G","G16","G16Y","G16Y10/00","G16Y10/80","G","G16","G16Y","G16Y20/00","G16Y20/20","G","G16","G16Y","G16Y40/00","G16Y40/10","G","G16","G16Y","G16Y40/00","G16Y40/20","G","G16","G16Y","G16Y40/00","G16Y40/50","H","H04","H04L","H04L1/00","H04L1/0001","H04L1/0002","H","H04","H04L","H04L1/00","H04L1/0001","H04L1/0006","H","H04","H04L","H04L41/00","H04L41/08","H04L41/0896","H","H04","H04L","H04L63/00","H04L63/04","H04L63/0428","H04L63/0435","H","H04","H04L","H04L67/00","H04L67/01","H04L67/10"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120950908A/en"},{"publication_number":"CN120951838A","title":"Initial value calculation method for pile composite foundation scheme based on BP neural network algorithm","abstract":"The invention discloses a pile-type composite foundation scheme initial value calculation method based on a BP neural network algorithm, which relates to the technical field of pile-type composite foundation design, and aims to obtain equal generation thickness and equal generation physical parameters of a soft soil layer, an upper soil layer or a lower soil layer at a calculation position of a to-be-determined scheme, obtain the embankment height at the calculation position of the to-be-determined scheme, input the obtained initial value of a design parameter in the pile-type composite foundation scheme at the calculation position of the to-be-determined scheme into the BP neural network after training is finished, set constraint conditions according to pile-type composite foundation specifications in the training process of the BP neural network, design an improved error performance function according to the constraint conditions, effectively avoid the problem that pile spacing parameters violate current specifications in the subsequent design parameter prediction, and promote the rationality of the BP neural network on the pile-type composite foundation design parameter initial value determination.","assignee":"Anhui Transport Consulting and Design Institute Co Ltd","inventors":["吴志刚","李翻翻","沈国栋","张志峰","杨大海","温广军","李明","张胜","郭城","陈星�"],"publication_date":"2025-11-14","filing_date":"2025-10-20","priority_date":"2025-10-20","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06F","G06F2111/00","G06F2111/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120951838A/en"},{"publication_number":"CN120950782A","title":"An automated data entry system and method based on a large model","abstract":"The invention discloses a data filling automation system and method based on a large model, which relate to the field of electric digital data processing and specifically comprise the steps of monitoring and collecting input behaviors of users in different platform forms in real time, constructing semantic state vectors, and aggregating to form semantic universe vectors; the method comprises the steps of constructing a structure disturbance tensor model, modeling the stable position of each field, obtaining a stable state representation vector of each target field under disturbance in a tensor projection mode, constructing a path cost function according to a matching relation between a semantic global vector and a structure mapping result, solving an optimal execution path of field filling, and driving a browser to automatically fill according to a path sequence to realize automatic operation of semantic reasoning and data filling under the structure disturbance. The method realizes automation and intellectualization of cross-platform data filling, effectively improves filling efficiency, has platform independence, semantic adaptability and structural fault tolerance, and ensures data consistency and stability of the filling process.","assignee":"Shandong Smile Data Technology Co ltd","inventors":["王方晓","张爱华","尹冲","潘若恒","王宁宁","王执祥"],"publication_date":"2025-11-14","filing_date":"2025-10-20","priority_date":"2025-10-20","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/958","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N5/00","G06N5/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120950782A/en"},{"publication_number":"CN120950269A","title":"A Deep Learning-Based Dynamic Allocation and Optimization Method and System for Computing Performance","abstract":"本发明公开了基于深度学习的算力性能动态分配优化方法及系统，包括：通过构建生成对抗网络与改进的Transformer模型结合架构，生成对抗网络的生成模型借助约束损失函数、任务依赖权重矩阵，生成贴合实际的初步算力分配方案；判别模型采用基于梯度的更新策略提升判别能力，改进的Transformer模型经位置编码动态调整、引入正则化项等优化，精准处理算力数据生成策略。系统中各单元协同，数据采集反馈单元实时获取算力数据，驱动模型周期性优化，分配指令转化单元结合波动补偿因子落实策略。最终实现算力资源的动态精准分配，本发明有效克服传统分配方式资源利用率低、响应迟缓的问题，显著提升算力性能与系统运行效率。 This invention discloses a method and system for dynamic allocation and optimization of computing power performance based on deep learning. The method includes: constructing a generative adversarial network (GAN) combined with an improved Transformer model; the GAN's generative model generates a preliminary computing power allocation scheme that fits reality by using a constraint loss function and a task-dependent weight matrix; the discriminative model employs a gradient-based update strategy to improve its discriminative ability; and the improved Transformer model is optimized through dynamic adjustment of position encoding and the introduction of regularization terms to accurately process the computing power data generation strategy. The system's various units collaborate: a data acquisition and feedback unit acquires computing power data in real time, driving periodic model optimization; and an allocation instruction conversion unit implements the strategy in conjunction with a fluctuation compensation factor. Ultimately, this achieves dynamic and accurate allocation of computing power resources. This invention effectively overcomes the problems of low resource utilization and slow response in traditional allocation methods, significantly improving computing power performance and system operating efficiency.","assignee":"Dalian Big Data Operations Co ltd","inventors":["高占普","高昊","陈剑"],"publication_date":"2025-11-14","filing_date":"2025-10-20","priority_date":"2025-10-20","cpc_codes":["G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5061","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5027","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5083","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","Y","Y02","Y02D","Y02D10/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120950269A/en"},{"publication_number":"CN120951182A","title":"A method and device for early warning of drilling spill risk","abstract":"本申请公开了一种钻井溢漏风险的预警方法及装置，涉及油气钻井勘探技术领域。本申请提供方案为：为按照预置时间间隔采集不同单位时刻的录井数据，并且根据在时间上排序先后将这些录井数据进行整合处理，以得到录井时序数据，由于录井时序数据本质上是时序信号，继而本申请对录井时序数据进行多尺度分解以提取出与溢流和井漏关联的一些关键的目标特征，目标特征包括时域特征、频域特征和时频域特征，然后再采用自注意力机制训练的风险预测模型对包含有这些目标特征的录井时序数据进行处理，目的是得到对溢流与井漏的风险预测结果。从而根据该风险预测结果判断是否触发风险预警。 This application discloses a method and apparatus for early warning of drilling spill risks, relating to the field of oil and gas drilling exploration technology. The proposed solution involves collecting logging data at different time intervals according to a preset time interval, and integrating these logging data based on their temporal order to obtain logging time-series data. Since logging time-series data is essentially a time-series signal, this application further decomposes the logging time-series data into multi-scale values to extract key target features associated with spills and well leakage. These target features include time-domain features, frequency-domain features, and time-frequency-domain features. Then, a risk prediction model trained using a self-attention mechanism is used to process the logging time-series data containing these target features, aiming to obtain risk prediction results for spills and well leakage. Based on these risk prediction results, a risk warning is then triggered.","assignee":"Xinjiang Tarim Petroleum Exploration And Development Headquarters Co ltd; Petrochina Co Ltd","inventors":["王清华","罗绪武","张重愿","李宁","祝兆鹏","陈龙","宋先知","叶明涛","陈志涛","史永哲"],"publication_date":"2025-11-14","filing_date":"2025-10-20","priority_date":"2025-10-20","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2431","E","E21","E21B","E21B47/00","E21B47/10","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2131","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06Q","G06Q50/00","G06Q50/02","G","G08","G08B","G08B29/00","G08B29/18","G08B29/185","G08B29/186","G","G08","G08B","G08B31/00","E","E21","E21B","E21B2200/00","E21B2200/22","G","G06","G06F","G06F2123/00","G06F2123/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120951182A/en"},{"publication_number":"CN120954737A","title":"Machine Learning-Based Methods, Systems, and Media for Prognostic Risk Analysis of Esophageal Cancer","abstract":"本发明公开了基于机器学习的食管癌预后风险分析方法、系统及介质，涉及人工智能技术与生物信息学技术领域，包括：步骤一、采集患者多模态数据并标准化处理，据以构建稳定特征集；步骤二、构建多种机器学习模型，基于稳定特征集建模实现各个机器学习模型的高低风险组预测，据以确定机器学习模型以及最优特征集；步骤三、计算最优特征集中各个特征之间的SHAP交互值，据以绘制交互值曲线，得到TopA个交互特征对，将交互特征对作为新构建的特征纳入步骤一的原始特征集，重复步骤一的特征筛选流程以及步骤二，从而得到最优机器学习模型，用于预测患者高低风险组；该预后风险分析方法对局部晚期食管癌患者高低风险组检测特异度大幅提升。 This invention discloses a machine learning-based method, system, and medium for esophageal cancer prognostic risk analysis, relating to the fields of artificial intelligence and bioinformatics. The method includes: Step 1, collecting and standardizing multimodal patient data to construct a stable feature set; Step 2, constructing multiple machine learning models, using the stable feature set to predict high and low risk groups for each model, thereby determining the machine learning model and the optimal feature set; Step 3, calculating the SHAP interaction values between features in the optimal feature set, plotting interaction value curves to obtain the Top A interaction feature pairs, incorporating these pairs as newly constructed features into the original feature set from Step 1, repeating the feature selection process from Step 1 and Step 2, thereby obtaining the optimal machine learning model for predicting high and low risk groups in patients. This prognostic risk analysis method significantly improves the specificity for detecting high and low risk groups in patients with locally advanced esophageal cancer.","assignee":"Hefei Institutes of Physical Science of CAS","inventors":["李雪玲","汪健涛","程旭","黄洁","朱婧怡"],"publication_date":"2025-11-14","filing_date":"2025-10-20","priority_date":"2025-10-20","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/30","G","G06","G06N","G06N20/00","G","G16","G16B","G16B25/00","G16B25/10","G","G16","G16B","G16B40/00","G","G16","G16H","G16H30/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120954737A/en"},{"publication_number":"CN118628733B","title":"Multi-behavior pattern collaborative training method for semi-supervised semantic segmentation training","abstract":"The invention discloses a multi-behavior mode collaborative training method for semi-supervised semantic segmentation training, which comprises the steps of learning a model R 1 and a model R 2 by using labeled images, generating pseudo labels for unlabeled data by each subnet, adding a behavior mode vector extraction module for each model, introducing a new data stream for each subnet, applying different characteristic disturbance on the data stream, and obtaining a total loss function of each model in training after integration.","assignee":"Minjiang University","inventors":["汪涛","李思恩"],"publication_date":"2025-11-14","filing_date":"2024-05-26","priority_date":"2024-05-26","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/20","G06V10/26","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN118628733B/en"},{"publication_number":"CN119443267B","title":"A method and system for solving mathematical problems based on feature classifier-based thought chain reasoning","abstract":"本发明公开了一种基于特征分类器的思维链推理数学问题求解方法及系统，包括：(1)获取数学问题集并选取正向与反向思维链示例进行拼接组成新集合；(2)将新集合元素分别输入模型进行推理构建单词级别的推理路径生成树，选择池化差值处理后的注意力权重矩阵作为节点特征进行存储；(3)遍历生成的所有推理路径生成树，筛选符合要求的节点构建特征分类器训练集；(4)使用支持向量机算法训练特征分类器；(5)通过训练好的特征分类器，参与预训练语言模型推理过程中路径的选择，获得较为准确的推理过程及答案。利用本发明，可以实现对于预训练语言模型推理路径更细颗粒度的调整把控，有利于其在数学问题上的准确推理求解，提升其泛化水平。 This invention discloses a method and system for solving mathematical problems based on thought chain reasoning using a feature classifier, comprising: (1) obtaining a set of mathematical problems and selecting forward and reverse thought chain examples to form a new set; (2) inputting the elements of the new set into the model to construct a word-level reasoning path generation tree, and selecting the attention weight matrix after pooling difference processing as the node feature for storage; (3) traversing all generated reasoning path generation trees, and selecting nodes that meet the requirements to construct a feature classifier training set; (4) training the feature classifier using the support vector machine algorithm; (5) using the trained feature classifier to participate in the path selection process of the pre-trained language model reasoning process, and obtaining a more accurate reasoning process and answer. Using this invention, a finer-grained adjustment and control of the reasoning path of the pre-trained language model can be achieved, which is beneficial for its accurate reasoning and solving of mathematical problems and improves its generalization level.","assignee":"Zhejiang University ZJU","inventors":["李嘉明","谢亮","王闻箫","林彬彬"],"publication_date":"2025-11-14","filing_date":"2024-10-23","priority_date":"2024-10-23","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06F","G06F17/00","G06F17/10","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2411"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN119443267B/en"},{"publication_number":"CN113807900B","title":"A Bayesian Optimization-Based RF Order Demand Forecasting Method","abstract":"本发明涉及一种基于贝叶斯优化的RF订单需求预测方法，用以实现同时对多种商品的未来需求或销量进行预测，包括以下步骤：1)对历史订单销量数据集进行数据预处理，并划分训练数据集和测试数据集作为基于随机森林的订单需求预测模型的输入量和输出量；2)通过贝叶斯优化方法获取订单需求预测模型的最优超参数，得到最优的订单需求预测模型；3)基于优化后的订单需求预测模型对订单商品未来的销量进行预测。与现有技术相比，本发明具有预测准确性高、快速寻优、多种时间序列预测等优点。 This invention relates to a Bayesian optimization-based Random Forest (RF) order demand forecasting method, used to simultaneously predict the future demand or sales volume of multiple commodities. The method includes the following steps: 1) Preprocessing the historical order sales dataset and dividing it into training and test datasets as input and output of a Random Forest-based order demand forecasting model; 2) Obtaining the optimal hyperparameters of the order demand forecasting model using a Bayesian optimization method, thus obtaining the optimal order demand forecasting model; 3) Predicting the future sales volume of the ordered commodities based on the optimized order demand forecasting model. Compared with existing technologies, this invention has advantages such as high prediction accuracy, rapid optimization, and the ability to forecast multiple time series data.","assignee":"East China University of Science and Technology","inventors":["严怀成","王潇","张皓","李郅辰","王孟","田永笑","陈辉","张长柱","王曰英","施开波"],"publication_date":"2025-11-14","filing_date":"2021-10-14","priority_date":"2021-10-14","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0202","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/24323","G","G06","G06N","G06N7/00","G06N7/01"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN113807900B/en"},{"publication_number":"CN120781945B","title":"Relation perception gating neural network link prediction method oriented to coal rock knowledge graph","abstract":"The invention relates to a coal-rock knowledge graph-oriented relation sensing gating neural network link prediction method, and belongs to the technical fields of knowledge graphs and artificial intelligence. The method comprises the steps of loading a coal rock knowledge graph data set, generating a lifting graph based on the knowledge graph data set and obtaining a lifting graph embedded representation, initializing relation embedded representation in the knowledge graph data set by utilizing the lifting graph embedded representation, updating entity representation by using an N-layer GIN graph neural network structure as a knowledge graph message transmission architecture, performing prediction reasoning by using the updated entity representation as a candidate entity distribution score, controlling information flow by a design relation sensing neuron RPRU module, balancing and combining local and global features to obtain more reasonable embedded representation of a prompt example, and adopting the N-layer GIN neural network structure to design a message transmission mechanism so as to more effectively capture graph structure information, thereby improving the effective utilization of graph prompt information in training and reasoning processes.","assignee":"Linyi University","inventors":["王星","朱仰瑞","姚双龙","陈吉","刘烨","杨亭","贾俊华","张问银","王海峰","刘志强"],"publication_date":"2025-11-14","filing_date":"2025-09-12","priority_date":"2025-09-12","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N5/00","G06N5/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120781945B/en"},{"publication_number":"CN120952082A","title":"Classifier-assisted neural network architecture search for multi-objective, multi-level learning optimization methods","abstract":"本发明公开了一种分类器辅助神经网络架构搜索多目标多层级学习优化方法，包括下述步骤：将神经网络架构搜索问题作为多目标优化问题进行建模，构建搜索空间；构建测试网络进行训练；通过非支配排序筛选帕累托最优解，基于帕累托前沿保留的算子，构建专用搜索空间；执行分类器辅助的基于层级学习的多目标进化优化算法，基于候选解筛选策略对进化后的更新的种群进行选择，基于筛选的解进行目标函数评估，更新档案库，迭代训练后输出档案库中的帕累托最优解，选取最后一代种群的最优帕累托前沿中的网络作为最优的神经网络架构。本发明通过分类器辅助的分层粒子群优化算法实现神经网络架构的高效搜索，得到满足分类准确性和推理延迟双目标的解决方案。 This invention discloses a classifier-assisted multi-objective, multi-level learning optimization method for searching neural network architectures, comprising the following steps: modeling the neural network architecture search problem as a multi-objective optimization problem and constructing a search space; constructing a test network for training; selecting Pareto optimal solutions through non-dominated sorting, and constructing a dedicated search space based on Pareto front preservation operators; executing a classifier-assisted hierarchical learning-based multi-objective evolutionary optimization algorithm, selecting from the updated population after evolution based on a candidate solution selection strategy, evaluating the objective function based on the selected solutions, updating the database, iteratively training and outputting the Pareto optimal solutions in the database, and selecting the network in the optimal Pareto front of the last generation population as the optimal neural network architecture. This invention achieves efficient search for neural network architectures through a classifier-assisted hierarchical particle swarm optimization algorithm, obtaining a solution that satisfies both classification accuracy and inference latency objectives.","assignee":"South China University of Technology SCUT","inventors":["陈伟能","董芳晨","魏凤凤"],"publication_date":"2025-11-14","filing_date":"2025-06-24","priority_date":"2025-06-24","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G06F18/2148","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/086"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120952082A/en"},{"publication_number":"CN114202122B","title":"A method for predicting urban traffic flow based on Markov clustering graph attention network","abstract":"The invention relates to an urban traffic flow prediction method based on a Markov clustering graph attention network, which comprises the following steps of 1, obtaining a time sequence flow matrix according to historical traffic travel data, 2, extracting natural structure information existing in a graph based on a Markov clustering algorithm idea to obtain a global correlation node matrix, 3, establishing and generating an antagonistic neural network model, wherein an improved graph attention module in a model generator does not limit neighbor nodes in first-order neighbor nodes like the graph attention network when acquiring space hidden features, but extends the neighbor nodes into global correlation node information obtained based on the Markov clustering algorithm, and a learning training model takes the learned model as an area traffic flow prediction model.","assignee":"Hebei Normal University","inventors":["魏志成","张韬毅","王玉波"],"publication_date":"2025-11-14","filing_date":"2021-12-13","priority_date":"2021-12-13","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q50/00","G06Q50/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN114202122B/en"},{"publication_number":"AU2025256165A1","title":"An advanced processing element and system","abstract":"A processing element for a quantum processing apparatus is disclosed. The processing element includes: a silicon substrate; a dielectric material, wherein the silicon substrate and the dielectric material form an interface; an electrode formed on the dielectric material for isolating one or more electrons in the silicon substrate to form a quantum dot; a group IV atom having a nuclear spin located in the wavefunction of the one or more electrons, the nuclear spin of the group IV atom entangled with the one or more electrons; and a control arrangement for controlling a quantum property of the quantum dot and/or the nuclear spin to operate as a qubit.","assignee":"Diraq Pty Ltd","inventors":["Andrew Dzurak","Bas Hensen","Wister Huang","Arne LAUCHT","Chih-Hwan Henry YANG"],"publication_date":"2025-11-13","filing_date":"2025-10-23","priority_date":"2019-01-31","cpc_codes":["B","B82","B82Y","B82Y10/00","G","G06","G06N","G06N10/00","G06N10/40","H","H10","H10D","H10D48/00","H10D48/383","H","H10","H10D","H10D48/00","H10D48/383","H10D48/3835","H","H10","H10P","H10P14/00","H10P14/40","H","H10","H10D","H10D48/00","H10D48/40"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025256165A1/en"},{"publication_number":"WO2025233706A1","title":"Method for improving accuracy of machine learning models","abstract":"Broadly speaking, embodiments of the present techniques provide a method for training neuro-symbolic ML models through partial label learning, to improve the accuracy of the predictions of the ML models. Advantageously, the present techniques provide a training method which improves how well the ML model is trained, by reducing the training problem to that of partial label learning. In other words, the present techniques train the neural module through partial label learning, rather than conventional multi-class learning techniques.","assignee":"Samsung Electronics Co Ltd","inventors":["Efthymia Tsamoura"],"publication_date":"2025-11-13","filing_date":"2025-04-03","priority_date":"2024-05-10","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N20/00","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"WO","kind":"application","source_url":"https://patents.google.com/patent/WO2025233706A1/en"},{"publication_number":"AU2024200653B2","title":"System and method for facilitating affective-state-based artificial intelligence","abstract":"In some embodiments, affective-state-based artificial intelligence may be facilitated. One or more growth or decay factors for a set of affective attributes of an artificial intelligence entity may be determined, and a set of affective values, which are associated with the set of affective attributes, may be continuously updated based on the growth or decay factors. An input may be obtained, and a response related to the input may be generated based on the continuously-updated set of affective values of the artificial intelligence entity. In some embodiments, the growth or decay factors may be updated based on the input and subsequent to the updating of the decay factors, the affective values may be updated based on the updated growth or decay factors.","assignee":"EmergeX LLC","inventors":["Roy FEINSON","Michael Joseph KARLIN","Ariel Mikhael KATZ"],"publication_date":"2025-11-13","filing_date":"2024-02-02","priority_date":"2018-01-29","cpc_codes":["G","G06","G06F","G06F40/00","G06F40/20","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/043","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/091","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N3/00","G06N3/12","G06N3/126","G","G06","G06N","G06N5/00","G06N5/02","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06N","G06N7/00","G06N7/02","G06N7/023"],"country":"AU","kind":"grant","source_url":"https://patents.google.com/patent/AU2024200653B2/en"},{"publication_number":"US20250348792A1","title":"Systems and Methods for Probabilistic Representation-Based Machine Learning","abstract":"An example method of providing inferences uses probabilistic machine learning trained without offline training. The method includes receiving a first set of inputs at a probabilistic machine-learning model. The model comprises a set of nodes sparsely coupled to one another in accordance with associations learned from a prior set of inputs. The method also includes generating, via a first subset of nodes, a first set of multi-dimensional vectors approximated by aggregating respective subsets of the first set of inputs. The method further includes generating, via a second subset of nodes, a second set of multi-dimensional vectors. The second set of vectors is approximated based on the first set of inputs and the first set of vectors. The method further includes generating an inference for the first set of inputs based on the second set of vectors.","assignee":"Scedastic Al Inc","inventors":["Kevin Sukmin Son"],"publication_date":"2025-11-13","filing_date":"2025-07-22","priority_date":"2024-01-21","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N7/00","G06N7/01"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250348792A1/en"},{"publication_number":"EP4647967A1","title":"Training generative artificial intelligence models","abstract":"A computer-implemented method for training generative artificial intelligence, AI, models is provided. The method includes providing, to a plurality of generative AI models, a constitution including a set of rules. The method further includes performing a plurality of iterative training steps for training the plurality of generative AI models. Each iterative training step includes assigning, to each model from among the plurality of generative AI models, a role from among a plurality of roles. The plurality of roles includes an actor and a judge. Each iterative training step further includes prompting the assigned actor model with an input, to generate content that complies with the constitution. Each iterative training step further includes prompting the assigned judge model with the content generated by the assigned actor model, to determine a likelihood of compliance that the content generated by the assigned actor model complies with the constitution. Each iterative training step further includes providing, to at least one model, a reward for training, using reinforcement learning, the at least one model. The reward is based on the likelihood of compliance determined by the assigned judge model. The roles are assigned to each model in the plurality of iterative training steps such that each of the plurality of generative AI models is assigned to each of the plurality of roles in at least one of the plurality of iterative training steps.","assignee":"Vodafone Group Services Ltd","inventors":["Oliver Mey"],"publication_date":"2025-11-12","filing_date":"2025-05-09","priority_date":"2024-05-09","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06N","G06N5/00","G06N5/04"],"country":"EP","kind":"application","source_url":"https://patents.google.com/patent/EP4647967A1/en"},{"publication_number":"KR20250159619A","title":"Method and apparatus for providing and managing sheet music including non-contact page-turner function","abstract":"본 발명의 일 실시예는 디스플레이 및 촬영 장치를 포함하는 전자 장치에 의해 수행되는 페이지터너 기능을 포함하는 악보 제공 및 관리 방법에 관한 것으로, 디스플레이에 악보에 포함된 제1 페이지가 표시 중 촬영 장치로부터 사용자의 동작을 포함하는 영상 정보를 획득하는 단계, 미리 학습된 동작 인식 모델을 이용하여 영상 정보로부터 사용자의 미리 설정된 넘김 요청 동작을 식별하는 단계, 넘김 요청 동작에 대응되는 페이지 넘김 요청 정보를 생성하는 단계 및 페이지 넘김 요청 정보를 기반으로 디스플레이에 제1 페이지를 악보에 포함된 다른 제2 페이지로 변경하여 표시하는 단계를 포함하는 것이 특징이다. One embodiment of the present invention relates to a method for providing and managing sheet music including a page-turner function performed by an electronic device including a display and a photographing device, the method comprising the steps of: obtaining image information including a user's motion from a photographing device while a first page included in the sheet music is displayed on a display; identifying a preset turning request motion of the user from the image information using a pre-learned motion recognition model; generating page-turning request information corresponding to the turning request motion; and changing the first page to a different second page included in the sheet music and displaying it on the display based on the page-turning request information.","assignee":"서덕식","inventors":["서덕식"],"publication_date":"2025-11-11","filing_date":"2025-10-23","priority_date":"2025-02-03","cpc_codes":["G","G10","G10H","G10H1/00","G10H1/0008","A","A47","A47B","A47B19/00","A47B19/002","G","G06","G06F","G06F3/00","G06F3/01","G06F3/017","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06V","G06V40/00","G06V40/20","G06V40/28","G","G09","G09B","G09B15/00","G09B15/02","G09B15/023","G","G10","G10G","G10G1/00","G","G10","G10H","G10H1/00","G10H1/36","G10H1/361","G10H1/368","G","G10","G10H","G10H2220/00","G10H2220/005","G10H2220/015"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250159619A/en"},{"publication_number":"CN120930760A","title":"Government knowledge graph body construction and optimization method, device, equipment and medium","abstract":"本发明公开了政务知识图谱本体构建及优化方法、装置、设备及介质，属于人工智能与知识工程交叉技术领域，本发明要解决的技术问题为如何为本体构建提供统一、规范的知识基准，实现系统自主感知本体变化特征，精准匹配最优分析模式，无需人工干预即可保障本体构建的高效性和准确性，技术方案为：构建领域知识库：启动本体构建系统，自动加载目标领域的基础配置参数，归集目标领域的核心知识要素及非结构化文档，并对非结构化文档进行分类、语义标注及格式统一处理，生成层级结构化的领域知识库，作为本体构建的统一知识基准；待处理数据获取及分析流程触发；自适应选择本体分析模式。 This invention discloses a method, apparatus, equipment, and medium for constructing and optimizing a government knowledge graph ontology, belonging to the interdisciplinary field of artificial intelligence and knowledge engineering. The technical problem this invention aims to solve is how to provide a unified and standardized knowledge benchmark for ontology construction, enabling the system to autonomously perceive ontology change characteristics, accurately match the optimal analysis mode, and ensure the efficiency and accuracy of ontology construction without manual intervention. The technical solution is as follows: Constructing a domain knowledge base: Starting the ontology construction system, automatically loading the basic configuration parameters of the target domain, collecting the core knowledge elements and unstructured documents of the target domain, and classifying, semantically annotating, and unifying the format of the unstructured documents to generate a hierarchical structured domain knowledge base as the unified knowledge benchmark for ontology construction; acquiring and triggering the data to be processed and analysis processes; and adaptively selecting the ontology analysis mode.","assignee":"Inspur Software Co Ltd","inventors":["张赫","陈晏鹏","盖家铭","司衍芹","张连超"],"publication_date":"2025-11-11","filing_date":"2025-10-16","priority_date":"2025-10-16","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/335","G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G","G06","G06F","G06F16/00","G06F16/30","G06F16/36","G06F16/367"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120930760A/en"},{"publication_number":"CN120932471A","title":"Automatic driving identification, signal timing and expressway ramp control method","abstract":"本发明涉及自动驾驶识别和混合交通管理技术领域，尤其是一种自动驾驶识别、信号机配时以及高速公路匝道控制方法。本发明首先获取交通监控数据，通过视频图像技术识别车辆的跟驰行为指标和换道行为指标；基于跟驰行为指标计算车辆的跟驰得分，如果跟驰得分大于设定的跟驰阈值，则判断车辆为自动驾驶车辆；如果跟驰得分小于或者等于跟驰阈值，则基于换道行为指标计算车辆的换道得分；换道得分大于设定的换道阈值，则判断车辆为自动驾驶车辆；反之，则判断车辆为人工驾驶车辆。基于车辆的群体行为特征进行CAV识别，不需要依赖车载设备，适用于所有车辆，提高了CAV识别的普适性和准确性；克服了现有技术中CAV识别依赖车载设备，难以保证识别准确性的缺陷。 This invention relates to the fields of autonomous driving recognition and hybrid traffic management technology, and in particular to a method for autonomous driving recognition, traffic signal timing, and highway ramp control. The invention first acquires traffic monitoring data and identifies vehicle following behavior indicators and lane-changing behavior indicators using video image technology. Based on the following behavior indicators, a following score is calculated for each vehicle. If the following score is greater than a set following threshold, the vehicle is determined to be an autonomous vehicle; if the following score is less than or equal to the following threshold, a lane-changing score is calculated based on the lane-changing behavior indicators; if the lane-changing score is greater than a set lane-changing threshold, the vehicle is determined to be an autonomous vehicle; otherwise, the vehicle is determined to be a manually driven vehicle. Based on the group behavior characteristics of vehicles, CAV recognition does not rely on onboard equipment and is applicable to all vehicles, improving the universality and accuracy of CAV recognition; it overcomes the shortcomings of existing CAV recognition technologies that rely on onboard equipment and are difficult to guarantee in terms of accuracy.","assignee":"Anhui University","inventors":["杨博","王飞飏","吴宇飞","汪奕倩","徐诺","胡智淋","李天赐","李文强"],"publication_date":"2025-11-11","filing_date":"2025-10-16","priority_date":"2025-10-16","cpc_codes":["G","G08","G08G","G08G1/00","G08G1/01","G08G1/017","G08G1/0175","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/251","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/004","G06N3/008","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06V","G06V20/00","G06V20/40","G06V20/46","G","G06","G06V","G06V20/00","G06V20/50","G06V20/52","G06V20/54","G","G08","G08G","G08G1/00","G08G1/07","G08G1/075","G","G08","G08G","G08G1/00","G08G1/07","G08G1/081","G","G06","G06V","G06V2201/00","G06V2201/07","G","G06","G06V","G06V2201/00","G06V2201/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120932471A/en"},{"publication_number":"CN120922180A","title":"Automatic driving vehicle dynamic environment modeling and active adaptation system","abstract":"The invention relates to automatic driving, in particular to an automatic driving vehicle dynamic environment modeling and active adaptation system, a multi-band cognitive radar module, a multi-band millimeter wave radar integrated with intelligent switching and cooperative working mechanisms, wherein the intelligent switching and cooperative working mechanisms are adopted to automatically optimize the duty cycle of the frequency band according to the environment conditions, transmit and receive radar waves of different frequency bands to acquire target multi-dimensional information, a data processing and dynamic environment modeling module is used for fusing the target multi-dimensional information acquired by the radar of different frequency bands by adopting a multi-sensor data fusion algorithm, the dynamic environment modeling is carried out on the surrounding of the vehicle, the dynamic environment model is updated in real time, and a decision and active adaptation control module is used for making reasonable driving decision and planning paths and carrying out active adaptation control.","assignee":"Anhui Falcon Wave Technology Co ltd","inventors":["胡宗品","李昂","路同亚","秦胜贤","刘志勇","任梦奇","李开文"],"publication_date":"2025-11-11","filing_date":"2025-10-16","priority_date":"2025-10-16","cpc_codes":["B","B60","B60W","B60W60/00","B60W60/001","B60W60/0015","B","B60","B60W","B60W10/00","B60W10/18","B","B60","B60W","B60W10/00","B60W10/20","B","B60","B60W","B60W40/00","B","B60","B60W","B60W50/00","B60W50/0098","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/251","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","B","B60","B60W","B60W2420/00","B60W2420/40","B60W2420/408","B","B60","B60W","B60W2710/00","B60W2710/18","B","B60","B60W","B60W2710/00","B60W2710/20","B","B60","B60W","B60W2720/00","B60W2720/10","Y","Y02","Y02T","Y02T10/00","Y02T10/10","Y02T10/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120922180A/en"},{"publication_number":"CN120931468A","title":"Watermark anti-counterfeiting evaluation method and system based on diffusion model, electronic equipment and storage medium","abstract":"The invention belongs to the technical field of watermark anti-counterfeiting evaluation, and provides a watermark anti-counterfeiting evaluation method, a system, electronic equipment and a storage medium based on a diffusion model, wherein the method comprises the steps of generating an auxiliary watermark data set, removing an image watermark, training an unconditional diffusion model, reversely forging watermark information, verifying counterfeit watermark and carrying out cluster analysis; the invention uses the unconditional diffusion model to simulate the watermark of the homologous non-watermark image, avoids the adaptability error caused by the manual selection data, improves the learning effect of the unconditional diffusion model on the target object, realizes the independent processing of suspicious watermark characteristics by using the unconditional diffusion model and the verification result of the target object as a model iteration loss function, improves the watermark removal effect, improves the watermark-forging resistance evaluation effect on the target object by reverse forging watermark, forging watermark verification and cluster analysis, and intuitively displays the potential watermark anti-counterfeiting problem.","assignee":"Hangzhou High Tech Zone Binjiang Blockchain And Data Security Research Institute; Zhejiang University ZJU","inventors":["董子平","程鹏","巴钟杰","王庆龙","任奎"],"publication_date":"2025-11-11","filing_date":"2025-10-16","priority_date":"2025-10-16","cpc_codes":["G","G06","G06T","G06T1/00","G06T1/0021","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/40","G06V10/54","G","G06","G06V","G06V10/00","G06V10/70","G06V10/762","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/95","G","G06","G06T","G06T2201/00","G06T2201/005","G06T2201/0065"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120931468A/en"},{"publication_number":"CN120921411A","title":"Robot electric control explosion-proof method and system for dual-mode pressure collaborative management","abstract":"本申请涉一种双模态压力协同管理的机器人电控防爆方法及系统，其包括通过获取机器人在运行周期内的环境信息并进行预处理；将标准化环境状态数据输入工况识别模型判断机器人处于正常工况或异常工况；在正常工况下执行节能型稳压控制策略，能在维持内部压力稳定的同时，通过调节电控系统输出功率实现节能效果，降低机器人的能耗；而在异常工况时启动高效泄压保护策略，可迅速降低内部压力，避免因压力过高引发爆炸等危险情况，保障机器人的安全运行；并且还利用深度学习算法分析历史运行数据并更新控制策略，能使控制策略不断优化；该方案可以实现双模态压力的协同管理，提升机器在不同工况下的适应性和防爆性能。 This application relates to a robot electronic control explosion-proof method and system with dual-modal pressure collaborative management. It includes: acquiring and preprocessing environmental information of the robot during its operating cycle; inputting standardized environmental state data into a working condition identification model to determine whether the robot is in a normal or abnormal working condition; implementing an energy-saving pressure stabilization control strategy under normal working conditions, which can maintain internal pressure stability while adjusting the output power of the electronic control system to achieve energy saving and reduce the robot's energy consumption; and activating an efficient pressure relief protection strategy under abnormal working conditions to quickly reduce internal pressure, preventing dangerous situations such as explosions caused by excessive pressure and ensuring the safe operation of the robot. Furthermore, it utilizes deep learning algorithms to analyze historical operating data and update the control strategy, enabling continuous optimization of the control strategy. This solution can achieve dual-modal pressure collaborative management, improving the robot's adaptability and explosion-proof performance under different working conditions.","assignee":"Beijing Yanling Jiaye Intelligent Technology Co ltd","inventors":["杨伟峰","任剑云","刘振宇","刘永鹏","李沁","杨海峰"],"publication_date":"2025-11-11","filing_date":"2025-10-16","priority_date":"2025-10-16","cpc_codes":["B","B25","B25J","B25J9/00","B25J9/16","B25J9/1674","B","B25","B25J","B25J9/00","B25J9/16","B25J9/1628","B25J9/163","B","B25","B25J","B25J9/00","B25J9/16","B25J9/1694","G","G05","G05B","G05B11/00","G05B11/01","G05B11/36","G05B11/42","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120921411A/en"},{"publication_number":"CN120931665A","title":"Method and system for analyzing peristomal skin based on image processing","abstract":"本申请涉及造口皮肤分析技术领域，其公开了一种基于图像处理的造口周围皮肤分析方法及系统，其首先获取包含造口、周围皮肤和部分造口袋底盘的原始图像，并进行预处理；接着，通过多目标语义分割精确分离出皮肤及底盘区域；然后，利用底盘作为机遇性参照物，通过比对其实际观测颜色与标准颜色，对图像进行色彩校准，以消除光照和设备差异。最后，在校准后的皮肤区域提取量化的颜色与纹理特征形成特征向量，并据此进行皮肤状态分析以生成报告。这样，无需额外色卡，通过利用图像内固有元素进行色彩校准，有效解决了居家环境下图像分析的色彩一致性难题，实现了对造口周围皮肤状态的自动化、客观化和标准化评估。 This application relates to the field of stoma skin analysis technology, and discloses a method and system for analyzing skin around the stoma based on image processing. First, it acquires and preprocesses an original image containing the stoma, surrounding skin, and part of the stoma bag base. Next, it accurately separates the skin and base regions through multi-objective semantic segmentation. Then, using the base as a chance reference, it performs color calibration on the image by comparing its actual observed color with a standard color to eliminate differences in lighting and equipment. Finally, it extracts quantified color and texture features from the calibrated skin region to form a feature vector, and uses this vector to perform skin condition analysis and generate a report. In this way, without the need for an additional color chart, it effectively solves the problem of color consistency in image analysis in a home environment by utilizing inherent elements within the image for color calibration, achieving automated, objective, and standardized assessment of the skin condition around the stoma.","assignee":"Nanjing University of Chinese Medicine","inventors":["曾钰","周娴","徐雯","洪艳燕","王会梅"],"publication_date":"2025-11-11","filing_date":"2025-10-16","priority_date":"2025-10-16","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06T","G06T5/00","G06T5/80","G","G06","G06T","G06T7/00","G06T7/40","G06T7/41","G","G06","G06V","G06V10/00","G06V10/20","G06V10/26","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G16","G16H","G16H30/00","G16H30/20","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30004","G06T2207/30088"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120931665A/en"},{"publication_number":"CN120932029A","title":"Image classification method and system based on multi-granularity mixed fine granularity prototype network","abstract":"The invention provides an image classification method and system based on a multi-granularity mixed fine-granularity prototype network, which are characterized in that an MIE module is used for extracting bottom fine-granularity features of different granularities from different layers of a Swin-transform backbone network of a visual model to be used as sources for corresponding type prototype learning, a prototype distillation strategy is used for refining the prototypes in a prototype library to enable the prototypes to accurately express the bottom features of objects, a prototype exchange strategy is used for enabling the visual model to learn to be more robust fine-granularity representation, namely, different types of prototypes are exchanged to enable the model to learn to be different in representation, prototype vector is finally matched with prototype representations corresponding to the prototypes of different types in the prototype library, and the prototype representations are in credible fusion with prediction results of a classification head of the visual model to obtain final predicted probability distribution, and classification accuracy and classification efficiency are improved.","assignee":"East China Jiaotong University","inventors":["余鹰","邓惟灏"],"publication_date":"2025-11-11","filing_date":"2025-10-16","priority_date":"2025-10-16","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N5/00","G06N5/04","G06N5/045","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/75","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/761","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120932029A/en"},{"publication_number":"CN113570029B","title":"Method for acquiring neural network model, image processing method and device","abstract":"本申请公开了人工智能领域中的一种获取神经网络模型的方法、图像处理方法及装置。其中，获取神经网络模型的方法包括：获取预训练的超网络模型，预训练的超网络模型是基于源数据集训练得到的；获取目标数据集，目标数据集对应的任务与源数据集对应的任务相同；基于目标数据集对预训练的超网络模型进行迁移学习，得到迁移学习后的超网络模型；在迁移学习后的超网络模型中搜索子网络模型，得到目标神经网络模型。本申请的方法能够在获得所需的神经网络模型的过程中降低训练成本，提高神经网络模型的性能。 This application discloses a method, image processing method, and apparatus for obtaining neural network models in the field of artificial intelligence. The method for obtaining the neural network model includes: obtaining a pre-trained supernetwork model, which is trained based on a source dataset; obtaining a target dataset, the task corresponding to the target dataset being the same as the task corresponding to the source dataset; performing transfer learning on the pre-trained supernetwork model based on the target dataset to obtain a transferred-learned supernetwork model; and searching for sub-network models in the transferred-learned supernetwork model to obtain the target neural network model. This method can reduce training costs and improve the performance of the neural network model during the process of obtaining the desired neural network model.","assignee":"Huawei Technologies Co Ltd","inventors":["田沈晶","黄泽毅","徐凯翔","唐少华"],"publication_date":"2025-11-11","filing_date":"2020-04-29","priority_date":"2020-04-29","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN113570029B/en"},{"publication_number":"US12468976B1","title":"Probabilistic inference in machine learning using a quantum oracle","abstract":"Methods, systems, and apparatus, including computer programs encoded on computer storage media, for using a quantum oracle to make inference in complex machine learning models that is capable of solving artificial intelligent problems. Input to the quantum oracle is derived from the training data and the model parameters, which maps at least part of the interactions of interconnected units of the model to the interactions of qubits in the quantum oracle. The output of the quantum oracle is used to determine values used to compute loss function values or loss function gradient values or both during a training process.","assignee":"Google LLC","inventors":["Nan Ding","Masoud MOHSENI","Hartmut Neven"],"publication_date":"2025-11-11","filing_date":"2021-05-27","priority_date":"2013-09-11","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06N","G06N10/00","G06N10/60","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N7/00","G06N7/01"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12468976B1/en"},{"publication_number":"US12468931B2","title":"Configuring a neural network using smoothing splines","abstract":"An embodiment includes identifying an initial plurality of sets of hyperparameter values at which to evaluate an objective function that relates hyperparameter values to performance values of a neural network. The embodiment also executes training processes on the neural network with the hyperparameters set to the each of the initial sets of hyperparameter values such that the training process provides an initial set of the performance values for the objective function. The embodiment also generates an approximation of the objective function using splines at selected performance values. The embodiment approximates a point at which the approximation of the objective function reaches a maximum value, then determines an updated set of hyperparameter values associated with the maximum value. The embodiment then executes a runtime process using the neural network with the hyperparameters set to the updated set of hyperparameter values.","assignee":"International Business Machines Corp","inventors":["Ulrich Alfons Finkler","Michele Merler","Mayoore Selvarasa JAISWAL","Hui Wu","Rameswar Panda","Wei Zhang"],"publication_date":"2025-11-11","filing_date":"2020-10-21","priority_date":"2020-10-21","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/211","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/217","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12468931B2/en"},{"publication_number":"CN113474231B","title":"Combined prediction and path planning for autonomous objects using neural networks","abstract":"传感器测量有关要操纵的对象(例如车辆或机器人)附近的行动者或其他对象的信息。传感器数据用于确定可操纵对象的可能动作序列，以实现确定的目标。对于要考虑的每个可能的动作，确定附近行动者或对象的一个或更多个可能的反应。在一些实施例中，这可以采用决策树的形式，具有对应于当前对象的可能动作和一个或更多个其他车辆或行动者的可能反应动作的交替等级的节点。机器学习可用于确定概率，以及沿着决策树的路径(包括序列)投影出选项。价值函数用于为每个考虑的序列或路径生成值，并且选择具有最高值的路径用于确定如何导航对象。 Sensors measure information about actors or other objects near the object to be manipulated (e.g., a vehicle or robot). Sensor data is used to determine possible sequences of actions for the manipulator to achieve a defined objective. For each possible action to be considered, one or more possible responses from nearby actors or objects are determined. In some embodiments, this can take the form of a decision tree with nodes at alternating levels corresponding to possible actions of the current object and one or more possible responses from other vehicles or actors. Machine learning can be used to determine probabilities and project options along paths (including sequences) of the decision tree. A value function is used to generate values for each considered sequence or path, and the path with the highest value is selected to determine how to navigate the object.","assignee":"Nvidia Corp","inventors":["B·达利","S·泰里","I·弗罗西奥","A·特罗科利"],"publication_date":"2025-11-11","filing_date":"2020-01-27","priority_date":"2019-02-05","cpc_codes":["G","G05","G05D","G05D1/00","G05D1/0088","B","B60","B60W","B60W60/00","B60W60/001","B","B60","B60W","B60W30/00","B60W30/08","B60W30/095","B","B60","B60W","B60W50/00","B60W50/0097","B","B60","B60W","B60W60/00","B60W60/001","B60W60/0011","B","B60","B60W","B60W60/00","B60W60/001","B60W60/0027","B60W60/00274","G","G05","G05D","G05D1/00","G05D1/02","G05D1/021","G05D1/0212","G05D1/0221","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N5/00","G06N5/01","G","G08","G08G","G08G1/00","G08G1/16","G08G1/161","B","B60","B60W","B60W2554/00","B60W2554/40","B60W2554/404","B60W2554/4041","B","B60","B60W","B60W2554/00","B60W2554/40","B60W2554/404","B60W2554/4042","B","B60","B60W","B60W2554/00","B60W2554/40","B60W2554/404","B60W2554/4044","B","B60","B60W","B60W2554/00","B60W2554/40","B60W2554/404","B60W2554/4045"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN113474231B/en"},{"publication_number":"CN113454648B","title":"Legend memory cells in recurrent neural networks","abstract":"训练具有使用勒让德存储器单元方程确定的连接权重的神经网络架构，同时可选地保持所确定的权重固定。网络可使用尖峰或非尖峰激活函数，可与其他神经网络架构堆叠或循环地耦合，并且可在软件和硬件中实现。本发明的实施方案提供了用于模式分类、数据表示和信号处理的系统，其使用跨越滑动时间窗口的正交多项式基函数进行计算。 A neural network architecture with connection weights determined using the Legendre memory cell equations is trained, while optionally keeping the determined weights fixed. The network can use spiked or non-spiculated activation functions, can be stacked or cyclically coupled with other neural network architectures, and can be implemented in software and hardware. Embodiments of the present invention provide systems for pattern classification, data representation, and signal processing that use orthogonal polynomial basis functions across a sliding time window for computation.","assignee":"Applied Brain Research Inc","inventors":["亚伦·R·沃克","克里斯托弗·大卫·以利亚史密斯"],"publication_date":"2025-11-11","filing_date":"2020-03-06","priority_date":"2019-03-06","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN113454648B/en"},{"publication_number":"CN120930921A","title":"Industrial system data resource aggregation and utilization platform system and method","abstract":"本发明公开了一种产业体系数据资源聚合利用平台系统和方法，属于智慧平台领域，包括多源异构数据融合中枢模块，用于基于向量空间映射机制，将企业生产数据、供应链数据、市场动态数据及政策法规数据转换为统一结构的图谱化数据模型，并通过图神经网络对各节点间的关联度进行动态权重重构；通过设置多源异构数据融合中枢模块，基于向量空间映射机制与图谱建模框架，将企业生产数据、供应链数据、市场动态数据与政策法规数据转化为统一结构的图谱化数据模型，并引入图神经网络对图谱节点间的关联度进行动态权重学习，从而有效突破了传统系统中存在的数据孤岛问题，提升了产业数据资源的聚合利用效率。 This invention discloses a platform system and method for aggregating and utilizing industrial system data resources, belonging to the field of intelligent platforms. It includes a multi-source heterogeneous data fusion central module, used to convert enterprise production data, supply chain data, market dynamics data, and policy and regulatory data into a unified structured graph-based data model based on a vector space mapping mechanism. A graph neural network is then used to dynamically reconstruct the weights of the relationships between nodes. By setting up this multi-source heterogeneous data fusion central module, based on a vector space mapping mechanism and a graph modeling framework, enterprise production data, supply chain data, market dynamics data, and policy and regulatory data are transformed into a unified structured graph-based data model. The introduction of a graph neural network to dynamically learn the weights of the relationships between graph nodes effectively overcomes the data silo problem existing in traditional systems and improves the efficiency of aggregating and utilizing industrial data resources.","assignee":"Huizhi Guoxing Beijing Technology Development Service Co ltd","inventors":["查保全","张雨阳","杨桢"],"publication_date":"2025-11-11","filing_date":"2025-07-22","priority_date":"2025-07-22","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G","G06","G06F","G06F16/00","G06F16/20","G06F16/25","G06F16/254","G","G06","G06F","G06F16/00","G06F16/20","G06F16/28","G06F16/284","G06F16/288","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/04","G06Q10/047","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120930921A/en"},{"publication_number":"CN120930078A","title":"Industrial park sewage quality parameter prediction method based on deep neural network","abstract":"本发明涉及工业园区污水水质预测技术领域，具体而言，涉及基于深度神经网络的工业园区污水水质参数预测方法，包括：获取工业园区污水排放口的多模态时间序列数据；构建工业园区污水水质参数预测模型；定义并执行基于遗传算法的自适应多目标优化算法，以求解工业园区污水水质参数预测模型的最佳参数配置，自适应多目标优化算法通过迭代进化，最大化用于评估模型综合性能的预设适应度函数的值；选取适应度函数值最高的模型参数配置作为最终预测模型；将工业园区污水水质参数预测模型部署于工业园区污水监控终端，接收实时待监测污水数据并输出工业园区水质参数的预测结果。本发明显著提升了工业园区污水水质参数预测的可靠性与决策透明度。 This invention relates to the field of industrial park wastewater quality prediction technology, specifically to a method for predicting industrial park wastewater quality parameters based on deep neural networks. The method includes: acquiring multimodal time-series data of wastewater discharge outlets in industrial parks; constructing a wastewater quality parameter prediction model for industrial parks; defining and executing an adaptive multi-objective optimization algorithm based on a genetic algorithm to solve for the optimal parameter configuration of the industrial park wastewater quality parameter prediction model. The adaptive multi-objective optimization algorithm iteratively evolves to maximize the value of a preset fitness function used to evaluate the overall performance of the model; selecting the model parameter configuration with the highest fitness function value as the final prediction model; and deploying the industrial park wastewater quality parameter prediction model on an industrial park wastewater monitoring terminal to receive real-time wastewater data to be monitored and output the prediction results of the industrial park water quality parameters. This invention significantly improves the reliability and decision-making transparency of industrial park wastewater quality parameter prediction.","assignee":"Leshan Normal University","inventors":["刘昶","孙永兴"],"publication_date":"2025-11-11","filing_date":"2025-10-13","priority_date":"2025-10-13","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/254","G06F18/256","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/12","G06N3/126","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/809","G06V10/811","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/10","G06V20/13","Y","Y02","Y02A","Y02A20/00","Y02A20/152"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120930078A/en"},{"publication_number":"CN120934875A","title":"Dynamic distribution method, system, equipment and storage medium for protection resources of power distribution network","abstract":"本发明公开了一种配电网防护资源动态分配方法、系统、设备及存储介质，涉及电力系统网络安全防护领域，方法包括：实时采集配电网的运行数据，构建多维度融合数据集，结合历史攻击模式库，通过攻击路径预测模型预测潜在攻击路径；对预测的潜在攻击路径进行风险评估，结合配电网拓扑结构和关键节点重要性，计算配电网各区域的防护资源需求量；根据风险评估结果，结合资源约束条件，实现防护资源的动态分配和实时调度；本发明能实时把握配电网运行，精准预测潜在攻击路径，准确评估风险并计算防护资源需求。实现防护资源动态分配与实时调度，合理利用资源，降低成本，提升防护效果，保障配电网在复杂网络攻击下安全稳定运行。 This invention discloses a method, system, device, and storage medium for dynamic allocation of protection resources in power distribution networks, relating to the field of power system network security protection. The method includes: real-time collection of distribution network operation data, construction of a multi-dimensional fusion dataset, and prediction of potential attack paths using an attack path prediction model, combined with a historical attack pattern database; risk assessment of the predicted potential attack paths, and calculation of protection resource requirements for each area of the distribution network, considering the distribution network topology and the importance of key nodes; and dynamic allocation and real-time scheduling of protection resources based on the risk assessment results and resource constraints. This invention can monitor distribution network operation in real time, accurately predict potential attack paths, accurately assess risks, and calculate protection resource requirements. It achieves dynamic allocation and real-time scheduling of protection resources, rationally utilizes resources, reduces costs, improves protection effectiveness, and ensures the safe and stable operation of the distribution network under complex network attacks.","assignee":"Guizhou Power Grid Co Ltd","inventors":["罗扶华","余云昊","张博达","文瑞彬","狄查美玲"],"publication_date":"2025-11-11","filing_date":"2025-08-29","priority_date":"2025-08-29","cpc_codes":["H","H04","H04L","H04L63/00","H04L63/14","H04L63/1408","H04L63/1416","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","H","H04","H04L","H04L41/00","H04L41/12","H","H04","H04L","H04L41/00","H04L41/14","H04L41/147","H","H04","H04L","H04L67/00","H04L67/01","H04L67/12","H","H04","H04L","H04L9/00","H04L9/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120934875A/en"},{"publication_number":"US12469322B2","title":"Methods and systems for transfer learning of deep learning model based on document similarity learning","abstract":"Disclosed is a method and system for transfer learning of a deep learning model based on a document similarity learning. A transfer learning method may include pre-training, by the at least one processor, a similarity model to output a similarity between documents, generating, by the at least one processor, a fine tuning model by replacing a first output function of the pre-trained similarity model with a second output function, and training, by the at least one processor, the fine tuning model to output a score for a document input to the fine tuning model.","assignee":"Naver Corp","inventors":["Sung Min Kim","Kyoungho CHOI","Kyuho LEE"],"publication_date":"2025-11-11","filing_date":"2021-06-23","priority_date":"2021-01-19","cpc_codes":["G","G06","G06V","G06V30/00","G06V30/40","G06V30/41","G06V30/418","G","G06","G06F","G06F17/00","G06F17/10","G06F17/16","G","G06","G06F","G06F17/00","G06F17/10","G06F17/18","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G06F18/2148","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/217","G06F18/2193","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06V","G06V30/00","G06V30/40","G06V30/41","G06V30/416"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12469322B2/en"},{"publication_number":"CN120929770A","title":"Decision information generation method and system based on multi-objective reinforcement learning","abstract":"The invention discloses a decision information generation method and system based on multi-objective reinforcement learning, which are applied to the field of computer information processing and comprise the steps of acquiring and utilizing a classification model to process multi-dimensional information flow data so as to determine importance and timeliness of information units, adopting a correlation analysis and clustering algorithm to generate an information subset which is correlated with preset key business elements and meets coverage, adopting a self-adaptive weight adjustment mechanism to update weights according to real-time change of business scenes and sort the information units, generating structured decision information through a decision template, and finally, carrying out parameter optimization on the classification model and the weight adjustment mechanism through a reinforcement learning feedback mechanism according to deviation of the decision information and actual business results.","assignee":"Shenzhen Youxun Cloud Computing Co ltd","inventors":["裘耀俊","刘家朝","温挺捷","刘中水"],"publication_date":"2025-11-11","filing_date":"2025-10-11","priority_date":"2025-10-11","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G","G06","G06F","G06F18/00","G06F18/20","G06F18/23","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120929770A/en"},{"publication_number":"KR20250159135A","title":"works, construction, management method using robot, drone, AI, VR, IoT, cloud, deep learning, metaverse","abstract":"본 발명은 a) 사람, 로봇, 드론, 아바타, 근로자, 작업자, 출입자, 장비, 거치대 등에 착용되는 수단, 필드수단, 웨어러블수단, 의류, 조끼, 헬멧, 밴드, 워치, 멜빵; b) 작업에 이용되는 드론, 장치, 장비, 로봇, 형틀, 비계, 굴착기, 펌프카, 크레인, 삼각대, 계측기; c) 상기 수단 내지 계측기 등에 구비되는 센서, 버튼, 랜턴, 카메라, 무전기, 알림기, 램프, 계측기, 배터리, 제어기, 오존기, 음이온기, 청정기, 히터, 쿨러; d) 상기 수단 내지 쿨러 등에 연계되는 비콘, 게이트웨이, 통신장치, 통신망, 서버, 처리장치, 단말기, 관제센터; e) 상기 수단 내지 관제센터 등에 이용되는 하드웨어, 소프트웨어, 어플리케이션; f) 상기 수단 내지 어플리케이션 등이 이용되는 작업, 업무, 제조, 생산, 공사, 건설, 건축, 관리, 안전관리, 자동관리, 출입관리, 공사관리, 작업관리, 환경관리, 건강관리, 냉난방관리; g) 상기 수단 내지 관리 등이 이용되는 공장, 가정, 현장, 건물, 사무실, 작업장, 산업현장; h) 상기 수단 내지 산업현장 등에 관련되는 방법, 공법, 시스템, 안전시스템, 작업시스템, 공사시스템, 무인시스템, 청정시스템; 등에 관한 것이다. The present invention relates to a) means worn by a person, a robot, a drone, an avatar, a worker, an operator, an entrant, equipment, a stand, a field means, a wearable means, clothing, a vest, a helmet, a band, a watch, a harness; b) a drone, a device, an equipment, a robot, a formwork, a scaffolding, an excavator, a pump car, a crane, a tripod, a measuring instrument used in a work; c) a sensor, a button, a lantern, a camera, a radio, an alarm, a lamp, a measuring instrument, a battery, a controller, an ozone generator, an anion generator, a purifier, a heater, a cooler provided in the means or the measuring instrument; d) a beacon, a gateway, a communication device, a communication network, a server, a processing device, a terminal, a control center connected to the means or the cooler; e) hardware, software, an application used in the means or the control center; f) a work, task, manufacturing, production, construction, architecture, management, safety management, automatic management, access control, construction management, work management, environmental management, health management, heating and cooling management, etc. in which the means or applications are used; g) Factories, homes, sites, buildings, offices, workshops, and industrial sites where the above means or management, etc. are used; h) Methods, construction methods, systems, safety systems, work systems, construction systems, unmanned systems, and cleaning systems related to the above means or industrial sites, etc.","assignee":"주식회사 한국산업기술원","inventors":["정하익"],"publication_date":"2025-11-10","filing_date":"2025-10-28","priority_date":"2022-10-24","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/08","B","B64","B64D","B64D1/00","B64D1/16","B64D1/18","B","B25","B25J","B25J11/00","B","B25","B25J","B25J13/00","B25J13/006","B","B25","B25J","B25J19/00","B25J19/02","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G08","G08B","G08B21/00","G08B21/02","G","G08","G08B","G08B25/00","G08B25/14","G","G08","G08C","G08C17/00","G08C17/02","H","H04","H04N","H04N21/00","H04N21/20","H04N21/21","H04N21/218","H04N21/2187","H","H04","H04N","H04N7/00","H04N7/18","B","B64","B64U","B64U2101/00","B64U2101/25","B64U2101/29","B","B64","B64U","B64U2101/00","B64U2101/45"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250159135A/en"},{"publication_number":"CO2025015157A2","title":"Model input based on the delay profile for ai/ml","abstract":"RESUMEN Se describen procedimientos y sistemas para el uso de un perfil de retardo como entrada para los modelos de AI/ML. La presente divulgación incluye soluciones para usar solamente un DP como entrada para un modelo de AI/ML, p. ej., donde el DP solo contiene el valor de temporización de múltiples trayectos recibidos. En comparación con el PDP, no se necesitan valores de potencia recibidos como entrada del modelo. Los resultados de las evaluaciones demuestran que el rendimiento del modelo de AI/ML se mantiene con solamente una pequeña degradación cuando se compara con el PDP usado como entrada del modelo. Un tamaño de la entrada del modelo sustancialmente reducido brinda una ventaja significativa a los modelos de AI/ML según la presente divulgación. Por ejemplo, la carga de mediciones se reduce significativamente. Este beneficio existe independientemente de si es necesario que las mediciones se señalicen de una entidad a otra entidad. Además, cuando es necesario que las mediciones de la entrada del modelo se envíen de una entidad a otra entidad, la sobrecarga de señalización se reduce significativamente. ABSTRACT This disclosure describes procedures and systems for using a delay profile as input for AI/ML models. It includes solutions for using only a delay profile (DP) as input for an AI/ML model, e.g., where the DP contains only the timing value of received multiple paths. Compared to a power-delay profile (PDP), no received power values are required as model input. Evaluation results demonstrate that AI/ML model performance is maintained with only minor degradation when compared to using a PDP as model input. A substantially reduced model input size provides a significant advantage to AI/ML models as described herein. For example, the measurement overhead is significantly reduced. This benefit exists regardless of whether measurements need to be signaled from one entity to another. Furthermore, when measurements from the model input need to be sent from one entity to another, the signaling overhead is significantly reduced.","assignee":"Ericsson Telefon Ab L M","inventors":["Jung-Fu Cheng","Yufei Blankenship"],"publication_date":"2025-11-07","filing_date":"2025-10-30","priority_date":"2023-04-07","cpc_codes":["H","H04","H04B","H04B17/00","H04B17/30","H04B17/309","H04B17/364","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N20/00"],"country":"CO","kind":"application","source_url":"https://patents.google.com/patent/CO2025015157A2/en"},{"publication_number":"KR20250158962A","title":"Apparatus and method for generating document images used in machine-learning of text detection and recognition","abstract":"본 발명은 학습 데이터의 생성 기술에 관한 것으로서, 상세하게는 이미지 내에 존재하는 텍스트를 검출 및 인식하는 모델을 학습할 때 학습 데이터로 사용되는 문서 이미지를 생성하기 위한 문서 이미지의 생성 장치 및 방법에 관한 것이다. 이를 위해, 본 발명에 따른 문서 이미지의 생성 방법은 컴퓨팅 장치에서 수행되는 문서 이미지의 생성 방법으로서, 정답 단어 박스 및 문자열 정보를 가진 문서 이미지를 입력받아 문서 이미지 내의 단어 박스의 위치를 검출하는 단계와, 상기 문서 이미지 내의 단어 박스에 있는 글자 영역을 배경 이미지로 생성하는 단계와, 상기 배경 이미지가 된 글자 영역에 임의의 폰트로 임의의 글자를 그려 새로운 문서 이미지를 생성하는 단계를 포함한다. The present invention relates to a technology for generating learning data, and more particularly, to a device and method for generating document images for generating document images used as learning data when training a model for detecting and recognizing text existing in an image. To this end, a method for generating a document image according to the present invention is a method for generating a document image performed in a computing device, comprising: a step of receiving a document image having a correct word box and string information and detecting the location of the word box within the document image; a step of generating a character area within the word box within the document image as a background image; and a step of drawing arbitrary characters in an arbitrary font in the character area that has become the background image to generate a new document image.","assignee":"주식회사 하나금융티아이","inventors":["문지영","여동훈","윤인용","김수현"],"publication_date":"2025-11-07","filing_date":"2025-10-22","priority_date":"2022-04-06","cpc_codes":["G","G06","G06V","G06V30/00","G06V30/10","G06V30/19","G06V30/191","G06V30/19147","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V30/00","G06V30/10","G06V30/20","G","G06","G06V","G06V30/00","G06V30/40","G06V30/41","G06V30/413"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250158962A/en"},{"publication_number":"CN120912906A","title":"Heliostat feature extraction method, heliostat pneumatic simulation method and heliostat pneumatic simulation equipment","abstract":"本发明公开了一种定日镜特征提取方法、定日镜气动仿真方法和设备，属于定日镜仿真技术领域。现有定日镜点云处理方法，缺乏特征挖掘手段，不能提取定日镜的几何特征，无法用于定日镜的气动仿真。本发明的一种定日镜特征提取方法，通过构建定日镜预处理模型、特征挖掘模型、特征融合模型，对定日镜点云扫描数据进行采样，得到定日镜关键点；然后对定日镜关键点进行特征挖掘，得到多维度几何特征，以捕捉定日镜的局部特征和全局特征；再对多维度几何特征进行非线性变换，得到结构特征量，并将其与环境特征量、定日镜仰角进行耦合，生成融合特征向量，从而可以对定日镜特征进行准确挖掘以及精准提取，进而可用于定日镜的气动仿真。 This invention discloses a heliostat feature extraction method, a heliostat aerodynamic simulation method, and equipment, belonging to the field of heliostat simulation technology. Existing heliostat point cloud processing methods lack feature mining techniques and cannot extract the geometric features of the heliostat, thus making them unsuitable for heliostat aerodynamic simulation. The heliostat feature extraction method of this invention constructs a heliostat preprocessing model, a feature mining model, and a feature fusion model. It samples the heliostat point cloud scanning data to obtain key points; then, it performs feature mining on these key points to obtain multi-dimensional geometric features, capturing both local and global features of the heliostat; finally, it performs nonlinear transformation on these multi-dimensional geometric features to obtain structural feature quantities, which are then coupled with environmental feature quantities and the heliostat elevation angle to generate a fused feature vector. This allows for accurate mining and precise extraction of heliostat features, which can then be used for heliostat aerodynamic simulation.","assignee":"Zhejiang Yuansuan Technology Co ltd","inventors":["郎超豪","吴健明"],"publication_date":"2025-11-07","filing_date":"2025-10-13","priority_date":"2025-10-13","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G","G06","G06F","G06F30/00","G06F30/20","G06F30/28","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/766","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/60","G06V20/64","G","G06","G06F","G06F2113/00","G06F2113/08","G","G06","G06F","G06F2119/00","G06F2119/14"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120912906A/en"},{"publication_number":"CN120915189A","title":"Intelligent motor state monitoring device and monitoring method","abstract":"本发明公开了一种智能电机状态监测装置和监测方法，包括处理器模块，处理器模块通过SPI通讯接口从电机驱动控制器获取永磁同步电机的电压电流信号和位置信号，在处理器模块中运行数据处理软件，数据处理软件结合解析模型和数据驱动的方式来进行电机状态监测和故障诊断，整个过程包括了特征提取、数据融合、故障分类和状态预测，最后把电机状态预测结果反馈到电机驱动控制系统进行容错控制。本发明的有益效果：利用专用处理器并结合数据处理手段来完成永磁同步电机状态监测和故障诊断，可快速完成电机状态的监控、故障诊断和状态预测，解决传统电机监测系统成本高、布线复杂、无法预测未来趋势的问题。 This invention discloses an intelligent motor condition monitoring device and method, including a processor module. The processor module acquires voltage, current, and position signals of a permanent magnet synchronous motor from a motor drive controller via an SPI communication interface. Data processing software runs within the processor module, combining analytical models and data-driven methods to perform motor condition monitoring and fault diagnosis. The entire process includes feature extraction, data fusion, fault classification, and condition prediction. Finally, the motor condition prediction results are fed back to the motor drive control system for fault-tolerant control. The advantages of this invention are: utilizing a dedicated processor combined with data processing techniques to complete permanent magnet synchronous motor condition monitoring and fault diagnosis, enabling rapid monitoring, fault diagnosis, and condition prediction of the motor condition, solving the problems of high cost, complex wiring, and inability to predict future trends in traditional motor monitoring systems.","assignee":"Yinuo Iot Technology Jiangsu Co ltd","inventors":["杜福嘉","赵峰","石晓勇"],"publication_date":"2025-11-07","filing_date":"2024-05-05","priority_date":"2024-05-05","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G01","G01R","G01R23/00","G01R23/16","G","G01","G01R","G01R31/00","G01R31/34","G01R31/343","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2411","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06N","G06N20/00","G06N20/10","H","H02","H02P","H02P21/00","H02P21/14","H","H02","H02P","H02P25/00","H02P25/02","H02P25/022","G","G06","G06F","G06F2218/00","G06F2218/08","G","G06","G06F","G06F2218/00","G06F2218/12","H","H02","H02P","H02P2207/00","H02P2207/05"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120915189A/en"},{"publication_number":"CN120744065B","title":"Group consensus large model illusion reducing method based on multi-model challenge","abstract":"本发明涉及一种基于多模型诘问的群体共识大模型幻觉降低方法，包括：根据多维综合评分结果从候选模型集中筛选出Top‑N模型，对Top‑N模型执行全组合双向知识蒸馏，得到初始模型群体，对初始模型群体中每个模型加载多个领域知识库进行领域自适应微调，构建得到群体模型集；给定用户问题，触发群体模型集中多个微调后的领域专家模型进行并行推理生成初始回答，通过构建诘问集、生成诘问指令以及更新回答进行迭代优化，在迭代优化过程中采用混合核函数计算群体回答的相似度，当相似度和稳定性同时达到预设阈值，或达到最大迭代次数时，终止迭代并输出结果。与现有技术相比，本发明具有回答准确性高以及领域适应性强等优点。 This invention relates to a method for reducing the illusion of large models in a group consensus based on multi-model questioning. The method includes: selecting Top-N models from a candidate model set based on multi-dimensional comprehensive scoring results; performing full-combination bidirectional knowledge distillation on the Top-N models to obtain an initial model group; loading multiple domain knowledge bases onto each model in the initial model group for domain-adaptive fine-tuning to construct a group model set; given a user question, triggering multiple fine-tuned domain expert models in the group model set to perform parallel reasoning to generate an initial answer; iterative optimization through constructing a question set, generating question instructions, and updating answers; calculating the similarity of group answers using a hybrid kernel function during the iterative optimization process; terminating the iteration and outputting the result when both similarity and stability reach a preset threshold, or when the maximum number of iterations is reached. Compared with existing technologies, this invention has advantages such as high answer accuracy and strong domain adaptability.","assignee":"Shanghai Academy of Agricultural Sciences; Shanghai Jiao Tong University","inventors":["焦杰然","张卫东","方献平","常丽英","孙志坚","胡智焕","贺世伟"],"publication_date":"2025-11-07","filing_date":"2025-08-20","priority_date":"2025-08-20","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3344","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3346","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120744065B/en"},{"publication_number":"CN119047512B","title":"Method and system for constructing large model for prediction and guidance of generated AI emotion propagation","abstract":"本发明提供一种基于生成式AI的情绪传播预测与引导大模型构建方法与系统，首先，构建情绪分析大模型，利用大语言模型和多源社交媒体数据，融合语法、语义和记忆驱动，采用稀疏门控混合专家训练技术，提升训练效率和性能；其次，基于情绪分析大模型开展舆情事件传播预测并且生成个体情绪与传播内容多维引导的受控制品，进行传播预测，同时生成情绪引导策略和内容；最后，构建综合系统进行演示验证，针对多样化突发舆情事件，基于时空特征分析网络舆情事件传播，进行实证分析，提高预测模型精度和效率。本发明突破了现有方法在处理复杂、多模态情绪表达和动态变化预测引导方面的局限性，有助于发现和疏导社会矛盾、维护社会稳定。 This invention provides a method and system for constructing a large-scale model for predicting and guiding the spread of emotions based on generative AI. First, a large-scale emotion analysis model is constructed, utilizing a large language model and multi-source social media data, integrating grammatical, semantic, and memory-driven approaches, and employing a sparse gating hybrid expert training technique to improve training efficiency and performance. Second, based on the emotion analysis model, the spread of public opinion events is predicted, and controlled objects guiding individual emotions and dissemination content in multiple dimensions are generated for propagation prediction, while simultaneously generating emotion guidance strategies and content. Finally, a comprehensive system is constructed for demonstration and verification. For diverse sudden public opinion events, the spread of online public opinion events is analyzed based on spatiotemporal characteristics, and empirical analysis is conducted to improve the accuracy and efficiency of the prediction model. This invention overcomes the limitations of existing methods in handling complex, multimodal emotion expressions and dynamically changing emotion prediction and guidance, contributing to the discovery and mitigation of social conflicts and the maintenance of social stability.","assignee":"Beijing University of Posts and Telecommunications","inventors":["陈健军","侯星宇"],"publication_date":"2025-11-07","filing_date":"2024-08-22","priority_date":"2024-08-22","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/951","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/24323","G","G06","G06F","G06F40/00","G06F40/20","G06F40/253","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098","G","G06","G06Q","G06Q10/00","G06Q10/40","Y","Y02","Y02T","Y02T10/00","Y02T10/10","Y02T10/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN119047512B/en"},{"publication_number":"CN120597173B","title":"An IoT Data Anomaly Detection Method Based on Directed Graph Spatiotemporal Feature Fusion","abstract":"本发明公开了基于有向图时空特征融合的物联网数据异常检测方法，包括以下步骤：S1、采集物联网中多个传感器节点的数据，构建传感器网络数据集；S2、对区域内的传感器网络数据进行清洗与预处理；S3、对处理后的数据通过贝叶斯变分推断学习传感器网络的有向图邻接矩阵，建模传感器节点间的有向因果关系；S4、对预测目标地点的传感器使用空间特征提取模块处理，利用有向图引导的图注意力机制提取多跳邻域空间特征；S5、对预测目标地点的传感器网络数据空间特征和有向图使用时空特征融合模块处理，动态调整有向图结构并进行数据预测；S6、生成异常得分和动态阈值并进行异常检测；该方法可提高物联网数据异常检测性能。 This invention discloses an anomaly detection method for IoT data based on directed graph spatiotemporal feature fusion, comprising the following steps: S1, collecting data from multiple sensor nodes in the IoT to construct a sensor network dataset; S2, cleaning and preprocessing the sensor network data within the region; S3, learning the directed graph adjacency matrix of the sensor network using Bayesian variational inference on the processed data to model the directed causal relationships between sensor nodes; S4, processing the sensors at the predicted target location using a spatial feature extraction module, and extracting multi-hop neighborhood spatial features using a directed graph-guided graph attention mechanism; S5, processing the spatial features and directed graph of the sensor network data at the predicted target location using a spatiotemporal feature fusion module, dynamically adjusting the directed graph structure and performing data prediction; S6, generating anomaly scores and dynamic thresholds and performing anomaly detection; this method can improve the performance of IoT data anomaly detection.","assignee":"Xiamen University","inventors":["肖珍龙","夏周翔"],"publication_date":"2025-11-07","filing_date":"2025-07-10","priority_date":"2025-07-10","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2433","G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06N","G06N7/00","G06N7/01"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120597173B/en"},{"publication_number":"CN120725120B","title":"Knowledge graph implementation methods, devices, equipment, and readable storage media","abstract":"The application provides a knowledge graph realization method, a device, equipment and a readable storage medium, wherein the method utilizes the appearance mode and appearance sequence of knowledge points in the analysis step of real questions, combines the joint modeling of bidirectional co-occurrence and unidirectional time sequence relation, automatically excavates actual use paths among the knowledge points from the analysis of the real questions, accurately captures implicit dependency and cognitive paths among the knowledge points, generates a dynamic knowledge graph with cognition rationality, improves the accuracy and rationality of knowledge association, enables the knowledge graph to be closer to real learning behavior, further converts the knowledge graph into a transfer matrix capable of predicting the grasping evolution of students to simulate the process of knowledge diffusion along a reasonable path, realizes the dynamic tracking and accurate prediction of the grasping state of the knowledge points of the students, and provides basis for personalized learning path recommendation.","assignee":"Beijing Century TAL Education Technology Co Ltd","inventors":["于娜","刘子韬","刘琼琼","牟宝奎","同庆"],"publication_date":"2025-11-07","filing_date":"2025-08-27","priority_date":"2025-08-27","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2458","G06F16/2465","G","G06","G06F","G06F16/00","G06F16/20","G06F16/28","G06F16/284","G06F16/288","G","G06","G06F","G06F16/00","G06F16/30","G06F16/36","G06F16/367","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120725120B/en"},{"publication_number":"CN120910345A","title":"User behavior sequence recommendation method and system based on graphic neural network","abstract":"The invention discloses a user behavior sequence recommending method based on a graph neural network, which comprises the steps of embedding a target user index, a historical interaction article sequence index and a target article index into a gate graph neural network by adopting an attention mechanism to perform message aggregation to respectively obtain a user vector, a historical interaction article vector and a target article vector, inputting the historical interaction article vector into a transducer encoder and a decoder structure to perform internal transmission, capturing a sequence dependency relationship to obtain an encoded article vector sequence, carrying out vector splicing on the encoded historical interaction article vector, the user vector and the target article vector after message aggregation, flattening to obtain a total vector sequence, transmitting the total vector sequence to a multi-layer perceptron neural network, carrying out linear transformation to obtain a scalar, carrying out nonlinear activation on the scalar, outputting whether the article is recommended to a user, and capturing the graph structure relationship and a time sequence behavior mode of the user-article.","assignee":"Shaanxi Normal University","inventors":["黄昭","陈嘉伟"],"publication_date":"2025-11-07","filing_date":"2025-07-14","priority_date":"2025-07-14","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G06F16/9535","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120910345A/en"},{"publication_number":"CN118391290B","title":"Primary fan running state intelligent monitoring method based on clustering association rule and ContiFormer network","abstract":"The invention discloses an intelligent monitoring method for the running state of a primary fan based on a clustering association rule and ContiFormer networks, which comprises the steps of firstly collecting a large amount of historical data generated in the running process of the primary fan, adopting an improved K-means clustering algorithm to perform clustering analysis on temperature, vibration, pressure and other data, searching frequent items through cyclic searching by using an Apriori algorithm, mining the association rule among the temperature, vibration, pressure and other data, introducing the association rule into a ContiFormer network model, and predicting a monitored variable by ContiFormer. And analyzing the prediction deviation by using a nuclear density estimation method, determining an early warning threshold value, and monitoring the running state of the primary fan so as to solve the problem of intelligent on-line monitoring of the running state of the primary fan when the unit is in flexible and changeable running conditions.","assignee":"North China Electric Power University","inventors":["黄从智","屈双艳"],"publication_date":"2025-11-07","filing_date":"2024-03-14","priority_date":"2024-03-14","cpc_codes":["F","F04","F04D","F04D27/00","F04D27/001","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2458","G06F16/2465","G","G06","G06F","G06F18/00","G06F18/20","G","G06","G06F","G06F18/00","G06F18/20","G06F18/23","G06F18/232","G06F18/2321","G06F18/23213","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06F","G06F2123/00","G06F2123/02","Y","Y02","Y02P","Y02P90/00","Y02P90/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN118391290B/en"},{"publication_number":"CN120911267A","title":"Transformer temperature rise assessment method and system based on Stacking integrated learning framework","abstract":"The invention discloses a transformer temperature rise assessment method and a transformer temperature rise assessment system based on a Stacking integrated learning framework, which adopt Latin hypercube sampling principle and CFD simulation, construct a sample data set based on input variables including oil baffle size, oil duct width, oil baffle number, boundary temperature and response variables including winding wire oil temperature rise and hot spot temperature rise, build a Stacking integrated temperature rise model integrating multi-model information in a layered mode, independently train a first layer by using the sample data set to generate a plurality of prediction results, train a second layer after splicing the prediction results to obtain the Stacking integrated temperature rise model integrating the multi-model information, and automatically complete training on the Stacking integrated temperature rise model integrating the multi-model information after super-parameter joint optimization determination, so as to obtain a trained temperature rise prediction model, and predict the temperature rise of a transformer to be predicted configured by a new oil baffle and an oil duct structure so as to obtain a predicted value of the output wire oil temperature rise and the hot spot value.","assignee":"Changzhou Xd Transformer Co ltd; Guangdong Power Grid Co Ltd; Shandong Power Equipment Co Ltd","inventors":["陈辛夫","肖威","史柏迪","郭鹏鸿","王新兵","徐莲环","钱艺华","王伟"],"publication_date":"2025-11-07","filing_date":"2025-07-22","priority_date":"2025-07-22","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06F","G06F30/00","G06F30/20","G06F30/28","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06F","G06F2111/00","G06F2111/10","G","G06","G06F","G06F2113/00","G06F2113/08","G","G06","G06F","G06F2119/00","G06F2119/08","Y","Y04","Y04S","Y04S10/00","Y04S10/50"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120911267A/en"},{"publication_number":"KR20250158729A","title":"Method, program, and apparatus for prediction of health status using electrocardiogram","abstract":"본 개시의 일 실시예에 따라 컴퓨팅 장치에 의해 수행되는, 심전도를 이용한 건강 상태의 예측 방법, 프로그램 및 장치가 개시된다. 상기 방법은, 사전 학습된 제 1 딥러닝 모델에 심전도 데이터를 입력하여, 상기 심전도 데이터로부터 특징을 추출하는 단계; 및 사전 학습된 제 2 딥러닝 모델에 상기 추출된 특징을 입력하여, 상기 심전도 데이터를 측정한 대상의 건강 상태를 예측하는 단계를 포함할 수 있다. 이때, 상기 제 1 딥러닝 모델은, 동일한 사람으로부터 측정된 제 1 학습 데이터 세트로부터 유사한 특징을 추출하도록 학습되고, 상이한 사람으로부터 측정된 제 2 학습 데이터 세트로부터 상이한 특징을 추출하도록 학습될 수 있다. According to one embodiment of the present disclosure, a method, program, and device for predicting a health condition using an electrocardiogram, which are performed by a computing device, are disclosed. The method may include a step of inputting electrocardiogram data into a first pre-trained deep learning model to extract features from the electrocardiogram data; and a step of inputting the extracted features into a second pre-trained deep learning model to predict the health condition of a subject whose electrocardiogram data has been measured. In this case, the first deep learning model may be trained to extract similar features from a first learning data set measured from the same person, and may be trained to extract different features from a second learning data set measured from different people.","assignee":"주식회사 메디컬에이아이","inventors":["권준명","조용연"],"publication_date":"2025-11-06","filing_date":"2025-10-30","priority_date":"2022-07-22","cpc_codes":["A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/318","A61B5/346","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/318","A61B5/346","A61B5/349","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7275","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/50","G","G16","G16H","G16H50/00","G16H50/70"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250158729A/en"},{"publication_number":"KR20250158724A","title":"System for recognizing character robust to various fonts","abstract":"AI(Artificial Intelligence) 기반 영상처리 기술을 통한 다양한 폰트에도 적용 가능한 문자 인식 시스템이 개시된다. 상기 시스템은, 문자 이미지를 입력받는 입력부, 상기 문자 이미지에서 문자 모양 정보(style)와 다양한 폰트 모양 정보(content)를 추출하는 추출부, 상기 추출된 다양한 폰트 모양 정보를 다양한 폰트에서 제거하여 하나의 문자 모양으로 변환하는 변환부, 및 상기 변환된 하나의 문자 모양을 이용하여 상기 문자 이미지에 대한 문자를 인식하는 인식부를 포함하는 것을 특징으로 한다. A character recognition system applicable to various fonts using AI (Artificial Intelligence)-based image processing technology is disclosed. The system is characterized by including an input unit for receiving a character image, an extraction unit for extracting character shape information (style) and various font shape information (content) from the character image, a conversion unit for removing the extracted various font shape information from various fonts and converting it into a single character shape, and a recognition unit for recognizing characters in the character image using the converted single character shape.","assignee":"한국전력공사","inventors":["서덕성","윤창걸","정세영"],"publication_date":"2025-11-06","filing_date":"2025-10-29","priority_date":"2022-08-12","cpc_codes":["G","G06","G06V","G06V30/00","G06V30/10","G06V30/24","G06V30/242","G06V30/244","G06V30/245","G","G06","G06N","G06N20/00","G","G06","G06V","G06V30/00","G06V30/10","G06V30/19","G06V30/191","G06V30/1914","G","G06","G06V","G06V30/00","G06V30/10","G06V30/19","G06V30/191","G06V30/19147"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250158724A/en"},{"publication_number":"KR20250158715A","title":"A kidney function prediction analysis method and a device using artificial intelligence learning model based on urine dipstick test information","abstract":"본 발명의 실시 예에 따른 방법은, 백신 부작용 예측 분석 장치의 동작 방법에 있어서, 백신 부작용 예측 분석 대상자의 대상자 변수 정보를 획득하는 단계; 상기 대상자 변수 정보에 대응하는 부작용 변수 정보를 획득하는 단계; 상기 대상자 변수 정보 및 상기 부작용 변수 정보를, 사전 구축된 백신 부작용 변수 학습 기반 인공지능 모델에 입력하여, 추정 백신 부작용 분류 모델 및 확률 정보를 획득하는 단계; 및 상기 추정 백신 부작용 분류 모델 및 확률 정보에 기초한 백신 부작용 예측 분석 정보를 출력하는 단계를 포함한다. A method according to an embodiment of the present invention comprises the steps of: obtaining subject variable information of a subject of vaccine side effect prediction analysis; obtaining side effect variable information corresponding to the subject variable information; inputting the subject variable information and the side effect variable information into a pre-built vaccine side effect variable learning-based artificial intelligence model to obtain an estimated vaccine side effect classification model and probability information; and outputting vaccine side effect prediction analysis information based on the estimated vaccine side effect classification model and probability information.","assignee":"차의과학대학교 산학협력단; 의료법인 성광의료재단","inventors":["장은찬","이채원","이상준","송규선","사순옥","홍명희","한현욱"],"publication_date":"2025-11-06","filing_date":"2025-10-22","priority_date":"2022-05-24","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/70","G","G06","G06N","G06N20/00","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H15/00","G","G16","G16H","G16H20/00","G16H20/10","G","G16","G16H","G16H40/00","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/50","G","G16","G16H","G16H50/00","G16H50/80"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250158715A/en"},{"publication_number":"AU2025252664A1","title":"Interactive basketball system","abstract":"Interactive basketball system Methods, systems, and apparatus, including computer programs encoded on computer storage media, for a basketball backboard. The basketball backboard includes a display screen, a plurality of sensors configured to generate sensor data regarding a shot attempt of a user, imaging devices configured to generate image data of the shot attempt, a speaker, and a control unit. The control unit can receive (i) the sensor data from the plurality of sensors and (ii) the image data from the imaging devices. Based on the received sensor data, the control unit can determine whether the shot attempt was successful. Based on the received image data and whether the shot attempt was successful, the control unit can generate analytics that indicate characteristics of the user and the shot attempt and recommendations for improving the shot attempt for subsequent shot attempts. The control unit can provide output data representing the analytics. Interactive basketball system","assignee":"Huupe Inc","inventors":["Paul Anton","Dan Hayes","Adam JABER","Matt MCCONAHA","Lyth SAEED"],"publication_date":"2025-11-06","filing_date":"2025-10-20","priority_date":"2020-08-19","cpc_codes":["A","A63","A63B","A63B24/00","A63B24/0021","A","A63","A63B","A63B24/00","A63B24/0062","A","A63","A63B","A63B24/00","A63B24/0084","A","A63","A63B","A63B63/00","A63B63/08","A63B63/083","A","A63","A63B","A63B69/00","A63B69/0071","A","A63","A63B","A63B71/00","A63B71/06","A63B71/0605","A","A63","A63B","A63B71/00","A63B71/06","A63B71/0619","A","A63","A63B","A63B71/00","A63B71/06","A63B71/0619","A63B71/0622","G","G01","G01S","G01S17/00","G01S17/88","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","A","A63","A63B","A63B24/00","A63B24/0021","A63B2024/0025","A","A63","A63B","A63B24/00","A63B24/0021","A63B2024/0028","A","A63","A63B","A63B24/00","A63B24/0021","A63B2024/0037","A","A63","A63B","A63B71/00","A63B71/06","A63B71/0619","A63B71/0622","A63B2071/0625","A","A63","A63B","A63B2220/00","A63B2220/05","A","A63","A63B","A63B2220/00","A63B2220/10","A63B2220/16","A","A63","A63B","A63B2220/00","A63B2220/17","A","A63","A63B","A63B2220/00","A63B2220/40","A","A63","A63B","A63B2220/00","A63B2220/62","A","A63","A63B","A63B2220/00","A63B2220/80","A63B2220/803","A","A63","A63B"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025252664A1/en"},{"publication_number":"KR20250158703A","title":"Apparatus for integrated analysis of medical images based on ai","abstract":"본 개시는 AI 기반 의료 영상 통합 분석 장치에 관한 것이다. 구체적으로, 본 개시의 따른 AI 기반 의료 영상 통합 분석 장치는, 서로 다른 모딜리티를 갖는 복수의 의료 영상을 통합하고, AI 기반의 병변 등급을 판정하며, 시각적 분석 근거를 포함하는 진단 리포트를 출력할 수 있다. The present disclosure relates to an AI-based medical image integration analysis device. Specifically, the AI-based medical image integration analysis device according to the present disclosure can integrate multiple medical images with different modalities, determine lesion grades based on AI, and output a diagnostic report including visual analysis evidence.","assignee":"임한솔","inventors":["임한솔"],"publication_date":"2025-11-06","filing_date":"2025-10-19","priority_date":"2025-10-19","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06T","G06T3/00","G06T3/40","G","G06","G06T","G06T5/00","G06T5/50","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G16","G16H","G16H15/00","G","G16","G16H","G16H30/00","G16H30/20","G","G16","G16H","G16H30/00","G16H30/40","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/50","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250158703A/en"},{"publication_number":"AU2025252601A1","title":"Integrated machine learning framework for optimizing unconventional resource development","abstract":"Implementations described and claimed herein provide systems and methods for developing resources from an unconventional reservoir. In one implementation, raw reservoir data for the unconventional reservoir is obtained. The raw reservoir data includes geology data, completion data, development data, and production data. The raw reservoir data is transformed to transformed data. The raw reservoir data is transformed to the transformed data based on a transformation from a set of one or more raw variable to a set of one or more transformed variables. The set of one or more transformed variables is statistically uncorrelated. Resource development data is extracted from the transformed data. Performance analytics are generated for the unconventional reservoir using the resource development data. The performance analytics are generated through ensemble machine learning. The unconventional reservoir is developed based on the performance analytics.","assignee":"ConocoPhillips Co","inventors":["Benjamin LASCAUD","Hui Zhou"],"publication_date":"2025-11-06","filing_date":"2025-10-17","priority_date":"2019-10-28","cpc_codes":["E","E21","E21B","E21B49/00","E21B49/08","E21B49/087","E","E21","E21B","E21B43/00","E","E21","E21B","E21B47/00","E21B47/003","E","E21","E21B","E21B7/00","E21B7/04","G","G01","G01V","G01V3/00","G01V3/18","G01V3/34","G","G06","G06N","G06N20/00","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G","G06","G06Q","G06Q50/00","G06Q50/02","E","E21","E21B","E21B2200/00","E21B2200/22","G","G01","G01V","G01V20/00","G","G01","G01V","G01V3/00","G01V3/38","G","G06","G06F","G06F17/00","G06F17/10","G06F17/18"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025252601A1/en"},{"publication_number":"AU2025252571A1","title":"Processing images using self-attention based neural networks","abstract":"Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing images using self-attention based neural networks. One of the methods includes obtaining one or more images comprising a plurality of pixels; determining, for each image of the one or more images, a plurality of image patches of the image, wherein each image patch comprises a different subset of the pixels of the image; processing, for each image of the one or more images, the corresponding plurality of image patches to generate an input sequence comprising a respective input element at each of a plurality of input positions, wherein a plurality of the input elements correspond to respective different image patches; and processing the input sequences using a neural network to generate a network output that characterizes the one or more images, wherein the neural network comprises one or more self-attention neural network layers.","assignee":"Google LLC","inventors":["Lucas Klaus BEYER","Mostafa DEGHANI","Alexey Dosovitskiy","Sylvain Gelly","Georg Heigold","Neil Matthew Tinmouth HOULSBY","Alexander KOLESNIKOV","Matthias Johannes Lorenz MINDERER","Thomas Unterthiner","Jakob D. Uszkoreit","Dirk WEISSENBORN","Xiaohua ZHAI"],"publication_date":"2025-11-06","filing_date":"2025-10-16","priority_date":"2020-10-02","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06T","G06T7/00","G06T7/97","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025252571A1/en"},{"publication_number":"US20250342372A1","title":"Concurrent running of inference workload instances on the same device resource using workload affinity","abstract":"A computer program product provides program instructions executable by a processor to cause the processor to perform operations. The operations include identifying a first inference workload instance using a first inference model and a second inference workload instance using a second inference model and identifying whether the first and second inference models have affinity for being run concurrently on a core processing unit. The first and second inference models have affinity if the first and second inference workload instances can run concurrently on the core processing unit without causing either of the first and second inference workload instances to experience latency above a predetermined limit. The operations further include causing the first and second inference workload instances to be run concurrently on the core processing unit if the first and second inference models have been identified to have affinity for being run concurrently on the core processing unit.","assignee":"Lenovo Enterprise Solutions Singapore Pte Ltd","inventors":["Tianyi Gao","Jiali Cheng","Jianfeng Guo","Jun Liu"],"publication_date":"2025-11-06","filing_date":"2024-05-05","priority_date":"2024-05-05","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250342372A1/en"},{"publication_number":"KR20250158692A","title":"Method and apparatus for determining class of image using ai model","abstract":"제1 분류기로부터 이미지의 클래스에 대한 제1 예측 결과를 수신하고, 제2 분류기로부터 이미지의 클래스에 대한 제2 예측 결과를 수신하는 단계, 제1 예측 결과 및 제2 예측 결과를 바탕으로 제1 AI 모델을 업데이트하는 단계, 그리고 업데이트된 제1 AI 모델을 사용하여 이미지의 클래스를 추론하는 단계를 통해 이미지의 클래스를 결정하는 방법 및 장치가 제공된다. A method and device for determining a class of an image are provided, including the steps of receiving a first prediction result for a class of an image from a first classifier, receiving a second prediction result for a class of the image from a second classifier, updating a first AI model based on the first prediction result and the second prediction result, and inferring the class of the image using the updated first AI model.","assignee":"삼성전자주식회사","inventors":["김병재","서성주","위삼 바다르","이희진","한승주"],"publication_date":"2025-11-06","filing_date":"2025-04-29","priority_date":"2024-04-30","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G06T7/001","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06V","G06V10/00","G06V10/40","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30108","G06T2207/30148"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250158692A/en"},{"publication_number":"US20250342404A1","title":"Simplistic machine learning model generation tool for predictive data analytics","abstract":"Systems and methods for predictive data analytics are provided. A method comprises generating a guided user interface (GUI) that guides one or more user operations on the user interface including: obtaining, from a database, a dataset including a plurality of data objects; determining one or more characteristics associated with a first data object of the plurality of data objects; identifying a subset of the dataset based at least in part on the one or more characteristics; selecting at least one machine learning algorithm; and training a machine learning (ML) model with respect to the first data object using the subset of the dataset and the at least one machine learning algorithm to generate a trained ML model; implementing the trained ML model with respect to the first data object in a cloud server to enable distributing the trained ML model to a plurality of client device via a network.","assignee":"State Farm Mutual Automobile Insurance Co","inventors":["Suresh B. Gajendran","Mahesh Chandrappa","Mark A. Dickneite","Charles T. Fiala","Rashid Zaheer"],"publication_date":"2025-11-06","filing_date":"2025-07-15","priority_date":"2020-08-13","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F8/00","G06F8/30","G06F8/34","G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/451","G06F9/453","G","G06","G06F","G06F40/00","G06F40/30","G06F40/35","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250342404A1/en"},{"publication_number":"US20250341809A1","title":"Action and/or process determination and recommendations for robotic process automation using semantic action graphs","abstract":"Action and/or process determination and recommendations for Robotic Process Automation (RPA) using semantic action graphs is disclosed. Semantic action graphs are graphs that store individual actions, and potentially graphical elements and/or text associated with the actions, as nodes, as well as the relationships between nodes as edges. Metadata to develop the semantic action graphs may be derived from task mining applications that can monitor the interactions of users with computing systems, workforce intelligence, etc. The semantic action graphs may be for a user, an organization, an industry, product-wide, etc. At their lowest level of granularity, the recommendations may be for mouse clicks, key presses, Application Programming Interface (API) calls, system events, etc. At higher levels of granularity, the recommendations may be for opening an order, creating a lead, approving a work item, etc.","assignee":"UiPath Inc","inventors":["Graham Sheldon","Yunjing Ma"],"publication_date":"2025-11-06","filing_date":"2024-05-01","priority_date":"2024-05-01","cpc_codes":["G","G06","G06N","G06N20/00","G","G05","G05B","G05B13/00","G05B13/02","G05B13/0265","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250341809A1/en"},{"publication_number":"KR20250158004A","title":"AI-Based Financial Information-Integrated Real Estate Transaction Registration and Risk Prediction System","abstract":"본 발명은 부동산 거래 시 금융정보·보증정보·등기정보·세금정보 등을 실시간으로 연동하여 거래대상물의 위험도를 인공지능(AI)을 이용해 자동으로 예측하는 금융정보 연동형 부동산 거래등록 및 위험도 예측 시스템에 관한 것이다. 본 시스템은 거래정보 수집모듈, 다기관 데이터 연동모듈, 인공지능 분석모듈, 결과처리모듈 및 사용자 인터페이스로 구성되며, 거래정보가 등록되는 즉시 인공지능 분석모듈이 자동 실행되어 거래금액, 근저당 설정비율, 보증보험 가입 여부, 체납이력, 시세변동률 등의 변수를 기반으로 시계열 분석 및 이상탐지 알고리즘을 통해 거래위험도를 산출한다. 산출된 위험도가 기준치를 초과할 경우 사용자 단말기 또는 행정기관에 경보가 전송되며, 거래정보와 분석결과는 블록체인 기반으로 저장되어 위·변조를 방지한다. 따라서 본 발명은 부동산 거래 과정에서 발생할 수 있는 전세사기나 불법전매 등의 위험을 사전에 탐지하고 예방함으로써 거래의 신뢰성과 투명성을 크게 향상시킬 수 있다. The present invention relates to a real estate transaction registration and risk prediction system linked to financial information, which automatically predicts the risk of a transaction object using artificial intelligence (AI) by linking financial information, guarantee information, registration information, tax information, etc. in real time during a real estate transaction. This system consists of a transaction information collection module, a multi-agency data linkage module, an artificial intelligence analysis module, a results processing module, and a user interface. As soon as transaction information is registered, the artificial intelligence analysis module automatically runs and calculates transaction risk through time series analysis and anomaly detection algorithms based on variables such as transaction amount, collateral setup ratio, insurance subscription status, delinquency history, and market fluctuation rate. If the calculated risk level exceeds the standard, an alert is sent to the user terminal or administrative agency, and transaction information and analysis results are stored on a blockchain basis to prevent forgery and falsification. Therefore, the present invention can significantly improve the reliability and transparency of transactions by detecting and preventing risks such as lease fraud and illegal resale that may occur during real estate transactions in advance.","assignee":"김충진","inventors":["김충진"],"publication_date":"2025-11-05","filing_date":"2025-10-19","priority_date":"2025-10-19","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/16","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q40/00","G06Q40/02","G06Q40/024","G","G06","G06Q","G06Q40/00","G06Q40/03","G06Q40/036","G","G06","G06Q","G06Q40/00","G06Q40/08","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26","H","H04","H04L","H04L9/00","H04L9/50"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250158004A/en"},{"publication_number":"KR20250157999A","title":"Command Processor, Neural processing system and Method for transmitting data thereof","abstract":"본 발명은 커맨드 프로세서, 뉴럴 프로세싱 시스템 및 그의 데이터 전송 방법을 개시한다. 상기 커맨드 프로세서는, 호스트 오프 칩 메모리를 포함하는 호스트 시스템으로부터 제1 도어벨을 수신하고, 상기 제1 도어벨에 따른 제1 워크로드 데이터를 수신하고, 상기 제1 워크로드 데이터를 분석하여 태스크 디스크립터 및 인터럽트 요청을 생성하는 워크로드 매니저, 상기 인터럽트 요청을 상기 호스트 시스템으로 전달하는 시스템 매니저 및 상기 태스크 디스크립터를 뉴럴 프로세서로 전달하고, 상기 태스크 디스크립터에 대한 리포트를 수신하는 뉴럴 프로세서 인터페이스를 포함하고, 상기 워크로드 매니저는, 상기 제1 도어벨에 따른 제1 버퍼 디스크립터를 로드하고, 상기 제1 버퍼 디스크립터에 따른 제1 커맨드 버퍼에 접근하여 상기 호스트 메모리로부터 상기 워크로드 데이터를 수신하고, 상기 제1 도어벨과 다른 제2 도어벨을 수신하고, 상기 제2 도어벨에 따라 커맨드 버퍼의 접근 없이 상기 호스트 오프 칩 메모리로부터 제2 워크로드 데이터를 수신한다. The present invention discloses a command processor, a neural processing system, and a data transmission method thereof. The command processor includes a workload manager that receives a first doorbell from a host system including a host off-chip memory, receives first workload data according to the first doorbell, analyzes the first workload data to generate a task descriptor and an interrupt request, a system manager that transmits the interrupt request to the host system, and a neural processor interface that transmits the task descriptor to a neural processor and receives a report on the task descriptor, wherein the workload manager loads a first buffer descriptor according to the first doorbell, accesses a first command buffer according to the first buffer descriptor, receives the workload data from the host memory, receives a second doorbell different from the first doorbell, and receives second workload data from the host off-chip memory according to the second doorbell without accessing a command buffer.","assignee":"리벨리온 주식회사","inventors":["김홍윤"],"publication_date":"2025-11-05","filing_date":"2025-10-17","priority_date":"2023-03-30","cpc_codes":["G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5061","G06F9/5066","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/461","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/48","G06F9/4806","G06F9/4812","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/48","G06F9/4806","G06F9/4843","G06F9/4881","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5027","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5061","G06F9/5077","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/52","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/54","G06F9/544","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06F","G06F2209/00","G06F2209/48","G06F2209/486","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/5017","G","G06","G06F","G06F2209/00","G06F2209/50","G06F2209/509"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250157999A/en"},{"publication_number":"EP4645325A1","title":"Method for generating medical reports","abstract":"Offenbart ist ein computer-implementiertes Verfahren zum Erzeugen von medizinischen Berichten, insbesondere Arztbriefen. Das Verfahren kann einen Schritt eines Empfanges von medizinischen Informationen über einen Patienten umfassen. Die empfangenen medizinischen Informationen können in natürlichsprachlichen Stichpunkten formuliert sein. Das Verfahren kann einen Schritt eines Erzeugens von natürlichsprachlichem Fließtext, welcher die empfangenen medizinischen Informationen widerspiegelt, für einen medizinischen Bericht durch ein trainiertes generatives Sprachmodell, umfassen. Revealbart is a computer-implemented method for generating medical reports, particularly medical letters. The method can include a step of receiving medical information about a patient. The received medical information may be formulated as natural language bullet points. The method can also include a step of generating natural language flow text, reflecting the received medical information, for a medical report using a trained generative language model.","assignee":"Myscribe GmbH","inventors":["Lars STOLL","Ira STOLL"],"publication_date":"2025-11-05","filing_date":"2024-05-04","priority_date":"2024-05-02","cpc_codes":["G","G16","G16H","G16H10/00","G16H10/60","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"EP","kind":"application","source_url":"https://patents.google.com/patent/EP4645325A1/en"},{"publication_number":"KR20250157338A","title":"System for managing renewable energy generator","abstract":"본 발명은 신재생에너지 발전 관리 시스템에 관한 것으로, 신재생에너지의 발전 실측치를 수집하는 복수 개의 RTU 및 상기 발전 실측치를 모니터링하여 발전 설비를 제어하는 분석서버를 포함하고, RTU는, 발전 실측치를 수집하여, 개별 인버터의 이상 여부를 판단하고, 이상 판단시 상기 분석서버로 알림을 전송하는 제1제어부 및 알림에 대한 상기 분석서버의 서버제어명령에 따라 발전 설비를 제어하되, 분석서버로부터 제어오류를 수신시 외부 통제센터로 전송하여 제어오류에 대응하여 기설정된 제어명령을 수신하여 제어대상을 제어하는 제2제어부를 포함하며, 분석서버는, 수신된 알림에 대한 조치율을 기반으로 상기 인버터의 이상 여부를 판단하는 기준값을 조정하는 조정값을 산출하여 해당 RTU로 전송할 수 있다. The present invention relates to a renewable energy generation management system, comprising a plurality of RTUs for collecting actual power generation data of renewable energy and an analysis server for monitoring the actual power generation data to control power generation facilities, wherein the RTUs include a first control unit for collecting actual power generation data, determining whether an individual inverter is abnormal, and transmitting a notification to the analysis server when an abnormality is determined, and a second control unit for controlling the power generation facilities according to a server control command of the analysis server in response to the notification, and for receiving a control error from the analysis server, transmitting the control error to an external control center, and receiving a preset control command in response to the control error to control a control target, and the analysis server can calculate an adjustment value for adjusting a reference value for determining whether the inverter is abnormal based on an action rate for the received notification, and transmit the calculated adjustment value to the corresponding RTU.","assignee":"주식회사 인코어드 테크놀로지스","inventors":["이효섭","임재륜","최상훈","이선정","김래균","박경남"],"publication_date":"2025-11-04","filing_date":"2025-10-29","priority_date":"2022-03-24","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/06","G","G06","G06N","G06N20/00","H","H02","H02J","H02J3/00","H02J3/004","H","H02","H02J","H02J3/00","H02J3/008","H","H02","H02J","H02J2101/00","H02J2101/20","H02J2101/22","H02J2101/24","H02J2300/24","Y","Y04","Y04S","Y04S10/00","Y04S10/12","Y04S10/123","Y","Y04","Y04S","Y04S50/00","Y04S50/10"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250157338A/en"},{"publication_number":"CN120893700A","title":"A Low-Carbon Control Method and System for the Cement Industry Based on Energy and Carbon Data","abstract":"The invention relates to the technical field of low-carbon control, in particular to a cement industry low-carbon control method and system based on energy-carbon data; firstly, determining a carbon emission boundary of a process industrial enterprise, determining a direct carbon emission source and an indirect carbon emission source, acquiring energy carbon data in real time by utilizing a carbon metering device, reporting the acquired energy carbon data to a carbon monitoring system by utilizing a carbon metering edge controller, acquiring characteristics and working condition data in the existing working condition system of the enterprise in real time, reporting the characteristics and working condition data to the carbon monitoring system, taking the energy carbon data, the characteristics and the working condition data as input by the carbon monitoring system, taking the optimal air-fuel ratio as output, performing fitting training by a stack self-coding neural network, learning by analyzing a large number of data samples to acquire a control rule, determining a parameter range reaching the optimal air-fuel ratio, and finally displaying a low-carbon control result by the carbon monitoring system, thereby realizing the effects of improving the decomposition rate of clinker and achieving energy conservation and carbon reduction.","assignee":"Zhejiang Wellsun Intelligent Technology Co Ltd","inventors":["何小龙","刘亚东","陆志荣","李梦娜","俞丹为","朱嘉琦","丁菊","吴潇琼","余建成"],"publication_date":"2025-11-04","filing_date":"2025-10-11","priority_date":"2025-10-11","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06Q","G06Q50/00","G06Q50/06"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120893700A/en"},{"publication_number":"CN120884256A","title":"Sleep staging method based on Markov chain dynamic loss","abstract":"本公开涉及基于马尔可夫链动态损失的睡眠分期方法，涉及数据处理领域。所述方法包括：对脑电信号进行预处理获取训练样本，构建并训练睡眠分期模型，利用训练好的模型进行睡眠分期，训练包括：计算基础分类损失；当训练样本睡眠阶段转移至不同阶段时，通过马尔可夫转移概率矩阵获得真实生理转移概率，若概率小于阈值，则根据模型是否正确预测了当前训练样本的睡眠阶段，计算训练样本的损失权重因子；根据损失权重因子与基础分类损失，计算序列感知损失的平均值；计算所述平均值对睡眠分期模型参数的梯度，并对睡眠分期模型的模型参数进行更新。本公开提升了睡眠分期模型对整体睡眠结构的理解和预测的生理学可解释性，进而提高了睡眠分期的准确性。 This disclosure relates to a sleep staging method based on Markov chain dynamic loss, belonging to the field of data processing. The method includes: preprocessing EEG signals to obtain training samples, constructing and training a sleep staging model, and using the trained model for sleep staging. Training includes: calculating a basic classification loss; when the training sample transitions to different sleep stages, obtaining the true physiological transition probability through a Markov transition probability matrix; if the probability is less than a threshold, calculating a loss weight factor for the training sample based on whether the model correctly predicted the current sleep stage; calculating the average value of the sequence perception loss based on the loss weight factor and the basic classification loss; calculating the gradient of the average value with respect to the sleep staging model parameters, and updating the model parameters of the sleep staging model. This disclosure improves the physiological interpretability of the sleep staging model in understanding and predicting the overall sleep structure, thereby improving the accuracy of sleep staging.","assignee":"Changchun University of Science and Technology","inventors":["李奇","张航","武岩","高宁","张安元"],"publication_date":"2025-11-04","filing_date":"2025-10-10","priority_date":"2025-10-10","cpc_codes":["A","A61","A61B","A61B5/00","A61B5/48","A61B5/4806","A61B5/4812","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/369","A61B5/372","A61B5/374","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4806","A61B5/4815","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120884256A/en"},{"publication_number":"US12462423B2","title":"Scene embedding for visual navigation","abstract":"Navigation instructions are determined using visual data or other sensory information. Individual frames can be extracted from video data, captured from passes through an environment, to generate a sequence of image frames. The frames are processed using a feature extractor to generate frame-specific feature vectors. Image triplets are generated, including a representative image frame (or corresponding feature vector), a similar image frame adjacent in the sequence, and a disparate image frame that is separated by a number of frames in the sequence. The embedding network is trained using the triplets. Image data for a current position and a target destination can then be provided as input to the trained embedding model, which outputs a navigation vector indicating a direction and distance over which the vehicle is to be navigated in the physical environment.","assignee":"Nvidia Corp","inventors":["Abel Karl Brown","Robert Stephen DiPietro"],"publication_date":"2025-11-04","filing_date":"2021-01-19","priority_date":"2018-08-13","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/70","G06T7/73","G06T7/74","G","G05","G05D","G05D1/00","G05D1/0088","G","G05","G05D","G05D1/00","G05D1/02","G05D1/021","G05D1/0212","G05D1/0221","G","G05","G05D","G05D1/00","G05D1/02","G05D1/021","G05D1/0231","G","G05","G05D","G05D1/00","G05D1/02","G05D1/021","G05D1/0231","G05D1/0246","G","G05","G05D","G05D1/00","G05D1/20","G05D1/24","G05D1/243","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06T","G06T7/00","G06T7/20","G06T7/246","G06T7/248","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/7715","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/10","G","G06","G06V","G06V20/00","G06V20/40","G06V20/46","G","G06","G06V","G06V20/00","G06V20/40","G06V20/49","G","G05","G05D","G05D2101/00","G05D2101/10","G05D2101/15","G","G06","G06N"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12462423B2/en"},{"publication_number":"CN119314673B","title":"A Method and System for Predicting Chronic Disease Risk in the Elderly Based on Temporal Knowledge Graphs","abstract":"The invention discloses a method and a system for predicting senile chronic disease risk based on a time sequence knowledge graph, belonging to the technical field of medical health information, wherein the method comprises the following steps of S1, acquiring a health data set of a senile crowd; the method comprises the steps of S2, based on a ESOA-XGBoost algorithm, carrying out feature selection and sequencing on each health data in a health data set to determine the correlation among variables and generate a health data feature set, S3, constructing and training a time sequence knowledge graph prediction model comprising a STKGR-PR model and a TIPNN model, and predicting the relation among a target entity, other entities and time sequence information through the time sequence knowledge graph prediction model, S4, predicting senile chronic disease risk of the old people according to the trained time sequence knowledge graph prediction model, and obtaining a senile chronic disease risk prediction result with good prediction effect and high accuracy through the method.","assignee":"Hainan University","inventors":["冯思玲","陈柏林","黄梦醒","王冠军","刘慧舟","徐博"],"publication_date":"2025-11-04","filing_date":"2024-09-30","priority_date":"2024-09-30","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/30","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N5/00","G06N5/02","Y","Y02","Y02A","Y02A90/00","Y02A90/10"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN119314673B/en"},{"publication_number":"US12462902B2","title":"Artificial intelligence engine architecture for generating candidate drugs","abstract":"A method is disclosed for using an artificial intelligence engine to generate candidate drug compounds, wherein the method comprises: generating candidate drug compounds comprising sequences via a creator module of the artificial intelligence engine. The method includes generating, via a descriptor module, a respective description for each of the candidate drug compounds at nodes in a knowledge graph, wherein the knowledge graph comprises a multi-dimensional representation of the candidate drug compounds and the respective description comprises drug compound structural information, drug compound activity information, and drug compound semantic information. The method includes determining a shape of the multi-dimensional representation of the candidate drug compounds; determining, based on the shape, a slice configured to be obtained from the representation; determining, using a decoder, which dimensions are included in the slice; and based on the dimensions, determining an effectiveness of a biomedical feature of the slice.","assignee":"Peptilogics Inc","inventors":["Francis Lee","Jonathan D. STECKBECK","Hannes Holste"],"publication_date":"2025-11-04","filing_date":"2021-06-04","priority_date":"2020-02-12","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G16","G16C","G16C20/00","G16C20/50","G","G16","G16C","G16C20/00","G16C20/70","G","G16","G16C","G16C60/00"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12462902B2/en"},{"publication_number":"US12462159B2","title":"Methods and systems for training a machine learning model with measurement data captured during manufacturing process","abstract":"Methods and systems for training a machine learning model with measurement data captured during a manufacturing process. Measurement data regarding a physical characteristic of a plurality of manufactured parts is received as measured by a plurality of sensors at various manufacturing stations. A time-series dynamics machine learning model encodes the measurement data into a latent space having a plurality of nodes. Each node is associated with the measurement data of one of the manufactured parts and at one of the manufacturing stations. A batch of the measurement data can be built, the batch include a first node and a first plurality of nodes immediately connected to the first node via first edges, and measured in time earlier than the first node. A prediction machine learning model can predict measurements of a first of the manufactured parts based on the latent space of the batch of nodes.","assignee":"Robert Bosch GmbH","inventors":["Filipe J. CABRITA CONDESSA","Devin T. WILLMOTT","Ivan BATALOV","João D. SEMEDO","Wan-Yi Lin","Jeremy KOLTER","Jeffrey Thompson"],"publication_date":"2025-11-04","filing_date":"2022-06-16","priority_date":"2022-06-16","cpc_codes":["G","G05","G05B","G05B19/00","G05B19/02","G05B19/418","G05B19/41885","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G05","G05B","G05B2219/00","G05B2219/30","G05B2219/32","G05B2219/32339","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12462159B2/en"},{"publication_number":"US12462075B2","title":"Resource prediction system for executing machine learning models","abstract":"A resource prediction system for executing machine learning models and method are provided. The system includes non-transitory memory storing instructions and a processor configured to execute the instructions to obtain input data including a targeted objective and the constraints, select a deployable machine learning model having an evaluation score that meets a predetermined criterion from among candidate machine learning models, virtually execute the deployable machine learning model on each of candidate hardware platforms according to the constraints, generate an assessment report of the virtual performance metrics set of the deployable machine learning model executed on each of the candidate hardware platforms, and select the suggested hardware platform meeting the predetermined criterion from among the candidate hardware platforms.","assignee":"Accenture Global Solutions Ltd","inventors":["Yao YANG","Andrew Hoonsik Nam","Mohamad Mehdi NASR-AZADANI","Teresa Sheausan Tung","Ophelia Min ZHU","Thien Quang NGUYEN","Zaid Tashman"],"publication_date":"2025-11-04","filing_date":"2021-02-23","priority_date":"2021-02-23","cpc_codes":["G","G06","G06F","G06F11/00","G06F11/30","G06F11/34","G06F11/3409","G","G06","G06F","G06F30/00","G06F30/20","G","G06","G06F","G06F11/00","G06F11/30","G06F11/3003","G06F11/302","G","G06","G06F","G06F11/00","G06F11/30","G06F11/34","G06F11/3457","G06F11/3461","G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N20/00","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06F","G06F11/00","G06F11/30","G06F11/32","G06F11/323","G","G06","G06F","G06F2111/00","G06F2111/04","G","G06","G06F","G06F2201/00","G06F2201/865"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12462075B2/en"},{"publication_number":"CN117441701B","title":"A UAV-based method and system for bird control in agriculture based on topological sorting reward mechanism","abstract":"The invention discloses an unmanned aerial vehicle agricultural bird repelling method and system based on a topological ordering rewarding mechanism, comprising the steps of constructing a bird recognizer, collecting three-dimensional target data of birds and carrying out three-dimensional target tracking; and using a reinforcement learning method of the reward mechanism based on topological ordering for unmanned aerial vehicle path planning, and using the unmanned aerial vehicle based on the reward mechanism of topological ordering for agricultural bird-driving work. The invention improves the parallelism and resource utilization rate of the system through a layered reinforcement learning idea based on the dependency relationship among tasks in the topological sorting, and applies the abstract task representation of the top layer generated by the system to the control of a continuous system of the bottom layer, thereby solving the actual problems of bird damage control, poor task coordination, easy path repeated coverage, high energy consumption, low efficiency and the like in the current agricultural bird-repellent field.","assignee":"Changzhou University","inventors":["朱晨阳","陈智宣","朱正伟","司雯","罗玮玢","朱金宇"],"publication_date":"2025-11-04","filing_date":"2023-10-25","priority_date":"2023-10-25","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","A","A01","A01M","A01M29/00","A01M29/06","A","A01","A01M","A01M29/00","A01M29/06","A01M29/10","A","A01","A01M","A01M29/00","A01M29/12","A","A01","A01M","A01M29/00","A01M29/16","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06Q","G06Q10/00","G06Q10/04","G06Q10/047","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","Y","Y02","Y02T","Y02T10/00","Y02T10/10","Y02T10/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN117441701B/en"},{"publication_number":"CN120893592A","title":"Nonlinear data generation and processing methods based on photons and photonic quantum computers","abstract":"The application provides a nonlinear data generation and processing method based on light quanta and a light quanta computer, wherein the method comprises the steps of obtaining nonlinear functions and attribute information of a nonlinear layer, constructing an initial light quanta line according to the frequency spectrum quantity and the attribute information of the nonlinear functions, measuring the initial light quanta line, determining an objective function corresponding to the initial light quanta line according to a measurement result, determining loss information of the objective function, determining whether the initial light quanta line is applied according to the loss information, taking the initial light quanta line as a target light quanta line if the initial light quanta line is applied, and calculating input data of the nonlinear layer based on the target light quanta line. The application can integrate the low energy consumption characteristic of photon calculation, the nonlinear characteristic of light quantum calculation and the general programmable capability of electronic calculation, and calculate the input data of the nonlinear layer with high energy efficiency. In addition, the method has the advantages of smaller complexity of the quantum wires and no participation of nonlinear optical devices.","assignee":"Shanghai Turing Intelligent Computing Quantum Technology Co Ltd","inventors":["朱宇泽","何俊杰","胡克明","杨林"],"publication_date":"2025-11-04","filing_date":"2025-10-10","priority_date":"2025-10-10","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/20","G","G06","G06N","G06N10/00","G06N10/40","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/067","G06N3/0675","Y","Y02","Y02D","Y02D10/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120893592A/en"},{"publication_number":"US12462144B2","title":"Domain generalization via batch normalization statistics","abstract":"Generally, the present disclosure is directed to systems and methods that leverage batch normalization statistics as a way to generalize across domains in particular, example implementations of the present disclosure can generate different representations for different domains by collecting independent batch normalization statistics, which can then be used to map between domains in a shared latent space. At test or inference time, samples from an unknown test or target domain can be projected into the same shared latent space. The domain of the target sample can therefore be expressed as a linear combination of the known ones, with the combination between weighted based on respective distances between batch normalization statistics in the latent space. This same mapping strategy can be applied at both training and test time to learn both a latent representation and a powerful but light-weight ensemble model that operates within such latent space.","assignee":"Google LLC","inventors":["Mattia Segù","Federico Tombari","Alessio Tonioni"],"publication_date":"2025-11-04","filing_date":"2021-03-05","priority_date":"2020-03-05","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12462144B2/en"},{"publication_number":"MX2025012508A","title":"An apparatus, a method and a computer program for video coding and decoding","abstract":"A method comprising: receiving input data comprising an initial reconstruction of an input data of an encoder, an encoded residual, wherein the encoded residual is determined based at least on a ground truth data of the encoder and the initial reconstruction of the input data and an adaptation signal into a decoder, said decoder comprising at least a first sub-decoder, a syntax generation unit and a probability model; obtaining an entropy decoded residual component based at least on the encoded residual component and on one or more probabilities of the probability model, wherein the entropy decoded residual is represented as a plurality of decoded latent tensor elements; providing the entropy decoded residual component to the first sub-decoder; providing the adaptation signal to the syntax generation unit for generating one or more syntaxes based on the adaptation signal; adapting decoding of the residual component in the first sub- decoder based on said one or more syntaxes; obtaining a reconstructed residual component from an output of the first sub-decoder; and combining the reconstructed residual component with initial reconstruction of the input data to obtain a final reconstruction of the input data.","assignee":"Nokia Technologies Oy","inventors":["Nannan Zou","Francesco Cricrì","Honglei Zhang"],"publication_date":"2025-11-03","filing_date":"2025-10-20","priority_date":"2023-04-25","cpc_codes":["H","H04","H04N","H04N19/00","H04N19/50","H04N19/503","H","H04","H04N","H04N19/00","H04N19/10","H04N19/134","H04N19/146","H04N19/147","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/13","H","H04","H04N","H04N19/00","H04N19/50","H04N19/593","H","H04","H04N","H04N19/00","H04N19/60","H04N19/61","H","H04","H04N","H04N19/00","H04N19/70"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2025012508A/en"},{"publication_number":"MX2025012420A","title":"Speculative decoding in autoregressive generative artificial intelligence models","abstract":"Certain aspects of the present disclosure provide techniques and apparatus for generating a response to a query input in a generative artificial intelligence model. An example method generally includes receiving a plurality of sets of tokens generated based on an input prompt and a first generative artificial intelligence model, each set of tokens in the plurality of sets of tokens corresponding to a candidate response to the input prompt; selecting, using a second generative artificial intelligence model and recursive adjustment of a target distribution associated with the received plurality of sets of tokens, a set of tokens from the plurality of sets of tokens; and outputting the selected set of tokens as a response to the input prompt.","assignee":"Qualcomm Inc","inventors":["Christopher Lott","Mingu Lee","Wonseok Jeon","Roland Memisevic"],"publication_date":"2025-11-03","filing_date":"2025-10-16","priority_date":"2023-04-20","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G06F16/9027","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2025012420A/en"},{"publication_number":"MX2025012419A","title":"Wireless communication method executed by means of authorized entity, and communication device","abstract":"Provided are a wireless communication method executed by means of an authorized entity, and a communication device. The method comprises: acquiring authorization information of a terminal device; determining, on the basis of the authorization information, whether to authorize an AI/ML service request related to the terminal device, wherein the AI/ML service request is used for requesting from the terminal device the transmission of information associated with the AI/ML service, and/or, the AI/ML service request is used for requesting that information of the terminal device is acquired to perform an AI/ML operation. In the present application, an authorization entity determines, on the basis of authorization information of a terminal device, whether to authorize an AI/ML service request related to a terminal device, such that an authorization mechanism related to the AI/ML service request is specified, thereby facilitating the prevention of an attacker from acquiring information related to an AI/ML service.","assignee":"Guangdong Oppo Mobile Telecommunications Corp Ltd","inventors":["Lihui Xiong","Jingran Chen"],"publication_date":"2025-11-03","filing_date":"2025-10-16","priority_date":"2023-12-25","cpc_codes":["H","H04","H04W","H04W12/00","H04W12/60","H04W12/63","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098","H","H04","H04L","H04L41/00","H04L41/16","H","H04","H04W","H04W12/00","H04W12/02","H","H04","H04W","H04W12/00","H04W12/06","H","H04","H04W","H04W12/00","H04W12/08","H","H04","H04W","H04W12/00","H04W12/60","H04W12/69","H04W12/72","H","H04","H04W","H04W12/00","H04W12/10","H04W12/104","H","H04","H04W","H04W12/00","H04W12/10","H04W12/108","H","H04","H04W","H04W84/00","H04W84/02","H04W84/04","H04W84/042"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2025012419A/en"},{"publication_number":"MX2025012296A","title":"Method for conditioning an intra frame codec on itself in an end-to-end learned codec","abstract":"A method, apparatus, and computer program product are provided. In the context of a method, the method receives, by a first codec, ground-truth data. The method, based on the ground-truth data, generates a first bitstream. The method, based on the first bitstream, generates initial reconstruction data, wherein the initial reconstruction data comprises a reconstruction of the ground-truth data. The method outputs, by the first codec, the initial reconstruction data. The method determines residual ground-truth data based at least on the initial reconstruction data and the ground-truth data. The method determines auxiliary data based at least on the first bitstream or based at least on data derived from the first bitstream. The method receives, by a second codec, the residual ground-truth data and the auxiliary data. The method, based on the residual ground-truth data and the auxiliary data, generates a second bitstream.","assignee":"Nokia Technologies Oy","inventors":["Nannan Zou","Francesco Cricrì","Honglei Zhang"],"publication_date":"2025-11-03","filing_date":"2025-10-14","priority_date":"2023-04-19","cpc_codes":["H","H04","H04N","H04N19/00","H04N19/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","H","H04","H04N","H04N19/00","H04N19/10","H04N19/102","H04N19/124","H","H04","H04N","H04N19/00","H04N19/90","H04N19/91"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2025012296A/en"},{"publication_number":"KR20250156037A","title":"TKE Droney drawings and chatracters","abstract":"TKE 조명방송드론을 그려 AI로 캐릭터화한 이미지 사용에 대한 특허권 또는 디자인권한을 사유화 Privatization of patent rights or design rights for the use of images of TKE's lighting broadcasting drones, which are AI-characterized.","assignee":"최성원","inventors":["최성원"],"publication_date":"2025-10-31","filing_date":"2025-10-14","priority_date":"2025-10-14","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250156037A/en"},{"publication_number":"CN120875189A","title":"Data center energy consumption optimization method and system based on artificial intelligence","abstract":"The application relates to the technical field of artificial intelligence, in particular to an energy consumption optimization method and system of a data center based on artificial intelligence, wherein the method comprises the steps of sequentially acquiring energy consumption data generated by the data center on each monitoring date according to a date sequence; the method comprises the steps of respectively checking a plurality of monitoring dates from a monitoring date interval through different date checking methods to serve as a plurality of key dates, generating different long-short-period memory network models based on energy consumption data of each date in the monitoring date interval and important numerical exercises corresponding to the energy consumption data according to each checking method, and acquiring the energy consumption data of the estimated date by using the recorded different long-short-period memory network models to execute energy consumption optimization processing. The application can accurately estimate the energy consumption data of the data center and timely perform energy consumption optimization processing.","assignee":"Ceicloud Data Storage Tech Beijing Co ltd","inventors":["孙鲁毅","孙茂金","孙明辉","周贤禹","孙伟"],"publication_date":"2025-10-31","filing_date":"2025-09-29","priority_date":"2025-09-29","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/24765","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06315","G","G06","G06Q","G06Q50/00","G06Q50/06","Y","Y02","Y02D","Y02D10/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120875189A/en"},{"publication_number":"CN120874028A","title":"Multi-tenant data leakage tracing method and system based on federal learning","abstract":"本申请涉及联邦学习技术领域，提供一种基于联邦学习的多租户数据泄露溯源方法和系统，采集多租户服务器的电磁辐射频谱数据，从中识别各服务器独有的电磁指纹特征和电磁辐射信号，基于电磁指纹特征以确定泄漏源位置坐标，将定位的电磁辐射信号转换为带有时空标记的目标频谱图形成电磁泄漏拓扑图，拓扑图与联邦全局模型聚合周期同步并利用水印触发器建立水印特征与电磁特征的动态关联关系，检测到关联异常，匹配异常水印特征与电磁拓扑图中的异常辐射点实现精准溯源。本申请解决了现有技术中联邦学习环境下针对物理侧信道攻击的泄露溯源能力低的问题，提升了联邦学习环境下针对物理侧信道攻击的泄露溯源能力。 This application relates to the field of federated learning technology, providing a method and system for tracing leaks in multi-tenant data based on federated learning. It collects electromagnetic radiation spectrum data from multi-tenant servers, identifies unique electromagnetic fingerprint features and electromagnetic radiation signals for each server, determines the location coordinates of the leak source based on the electromagnetic fingerprint features, and converts the located electromagnetic radiation signals into a target spectrum map with spatiotemporal markers to form an electromagnetic leak topology map. The topology map is periodically synchronized with the federated global model aggregation cycle, and a dynamic correlation between watermark features and electromagnetic features is established using a watermark trigger. When an anomaly is detected, the abnormal watermark features are matched with abnormal radiation points in the electromagnetic topology map to achieve accurate source tracing. This application solves the problem of low leak tracing capability against physical side-channel attacks in the existing technology under federated learning environment, and improves the leak tracing capability against physical side-channel attacks under federated learning environment.","assignee":"Ningbo Zihe Technology Co ltd","inventors":["门嘉平"],"publication_date":"2025-10-31","filing_date":"2025-09-29","priority_date":"2025-09-29","cpc_codes":["H","H04","H04L","H04L9/00","H04L9/002","H04L9/003","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F21/00","G06F21/10","G06F21/16","G","G06","G06F","G06F21/00","G06F21/30","G06F21/31","G06F21/32","G","G06","G06N","G06N20/00","H","H04","H04L","H04L9/00","H04L9/12","H","H04","H04L","H04L9/00","H04L9/32","H04L9/3297","H","H04","H04L","H04L2463/00","H04L2463/146"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120874028A/en"},{"publication_number":"CN120872680A","title":"Cloud-protogenesis-based automatic fault injection and recovery testing method and system","abstract":"本发明属于云计算与软件测试技术领域，具体公开了一种基于云原生的自动化故障注入与恢复测试方法，包括获取待测试云原生系统的测试基础数据；根据测试基础数据，构建故障测试执行模型；根据故障测试执行模型，执行自动化故障注入操作；根据故障注入操作之后的系统运行状态，生成异常告警日志，并同步记录故障注入操作日志；根据异常告警日志，执行自动化故障恢复操作；根据故障注入操作日志、系统监控数据及恢复操作过程日志，生成故障测试报告。本发明旨在解决现有技术在云原生系统故障测试中自动化程度低、场景覆盖窄、监控评估弱以及恢复被动的问题，提供一种全面化、自动化、可监控、可评估的故障注入与恢复测试方案。 This invention belongs to the field of cloud computing and software testing technology, specifically disclosing an automated fault injection and recovery testing method based on cloud-native computing. The method includes: acquiring basic test data of the cloud-native system under test; constructing a fault test execution model based on the basic test data; executing automated fault injection operations based on the fault test execution model; generating anomaly alarm logs based on the system's operating status after the fault injection operations, and simultaneously recording the fault injection operation logs; executing automated fault recovery operations based on the anomaly alarm logs; and generating a fault test report based on the fault injection operation logs, system monitoring data, and recovery operation process logs. This invention aims to solve the problems of low automation, narrow scenario coverage, weak monitoring and evaluation, and passive recovery in existing technologies for cloud-native system fault testing, providing a comprehensive, automated, monitorable, and evaluable fault injection and recovery testing solution.","assignee":"Happy Mutual Entertainment Shanghai Technology Co ltd","inventors":["赵高"],"publication_date":"2025-10-31","filing_date":"2025-09-29","priority_date":"2025-09-29","cpc_codes":["G","G06","G06F","G06F11/00","G06F11/07","G06F11/0703","G06F11/079","G","G06","G06F","G06F11/00","G06F11/07","G06F11/0703","G06F11/0793","G","G06","G06F","G06F11/00","G06F11/22","G06F11/2273","G","G06","G06F","G06F11/00","G06F11/30","G06F11/3065","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120872680A/en"},{"publication_number":"CN120876264A","title":"Infrared and visible light image enhancement fusion method and system for low-illumination scene","abstract":"本发明提出一种低照度场景的红外与可见光图像增强融合方法及系统，该方法包括：获取红外图像与可见光图像，对红外图像与可见光图像进行预处理；将亮度图像和光照先验图像输入至编码器中，通过最小化重建损失函数进行单模态重建预训练；对双通道可见光图像和归一化后的红外图像利用预训练的编码器进行深层特征提取，采用跨模态交叉注意力机制，通过交换查询向量的方式在融合模块中进行双向特征引导与融合，对融合后的特征利用预训练的解码器进行重建；本发明将低照度图像增强与跨模态特征融合集成于统一架构，通过端到端联合建模实现亮度提升、细节保留与模态一致性同步优化，避免了增强与融合流程割裂造成的信息损失。 This invention proposes a method and system for enhancing and fusing infrared and visible light images in low-light scenes. The method includes: acquiring infrared and visible light images and preprocessing them; inputting brightness images and prior illumination images into an encoder and performing single-modal reconstruction pre-training by minimizing the reconstruction loss function; extracting deep features from the dual-channel visible light image and the normalized infrared image using the pre-trained encoder; employing a cross-modal cross-attention mechanism to perform bidirectional feature guidance and fusion in the fusion module by exchanging query vectors; and reconstructing the fused features using a pre-trained decoder. This invention integrates low-light image enhancement and cross-modal feature fusion into a unified architecture, achieving simultaneous optimization of brightness enhancement, detail preservation, and modal consistency through end-to-end joint modeling, thus avoiding information loss caused by the separation of enhancement and fusion processes.","assignee":"Jiangxi University of Finance and Economics","inventors":["化定丽","陈庆茂","吴之亮","方玉明","左一帆","李强"],"publication_date":"2025-10-31","filing_date":"2025-09-29","priority_date":"2025-09-29","cpc_codes":["G","G06","G06T","G06T5/00","G06T5/50","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T5/00","G06T5/60","G","G06","G06T","G06T5/00","G06T5/90","G","G06","G06T","G06T7/00","G06T7/40","G","G06","G06T","G06T7/00","G06T7/90","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10004","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10048","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20212","G06T2207/20221","Y","Y02","Y02T","Y02T10/00","Y02T10/10","Y02T10/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120876264A/en"},{"publication_number":"CN120870232A","title":"Flux uniformity real-time evaluation method in flux smelting production","abstract":"The invention discloses a method for evaluating flux uniformity in smelting flux production in real time, which relates to the technical field of data processing and comprises the steps of constructing melt solidification resistance layer distribution in particles in a thermal radiation image, identifying component segregation boundaries in flux particles according to the melt solidification resistance layer distribution, generating segregation position marks based on the component segregation boundaries and the thermal radiation image, verifying time sequence correspondence between the segregation position marks and microstructure evolution processes of the flux particles, establishing a causal chain, adjusting the melt solidification resistance layer distribution according to recapture of heat conduction shunt in the thermal radiation image when chain breakage occurs in the causal chain, determining the component uniformity of the flux particles based on the adjusted melt solidification resistance layer distribution, and realizing high-precision evaluation of the component uniformity of the flux particles by monitoring the thermal radiation characteristics of the flux particles and accurately analyzing the dynamic evolution of the melt solidification resistance layer and the component segregation.","assignee":"Hunan Dongan Xiangjiang Technology Co ltd","inventors":["赵立基"],"publication_date":"2025-10-31","filing_date":"2025-09-29","priority_date":"2025-09-29","cpc_codes":["G","G01","G01N","G01N25/00","G01N25/20","G","G01","G01N","G01N15/00","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10048","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30108","G06T2207/30152","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30232"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120870232A/en"},{"publication_number":"CN120877071A","title":"Pathological image analysis model training method, device, equipment and storage medium","abstract":"The invention discloses a pathological image analysis model training method, a device, equipment and a storage medium, wherein the method comprises the steps of constructing an initial pathological image analysis model, wherein the initial pathological image analysis model comprises a sharing aggregator, a first expert network, a second expert network and a consistency constraint unit; the method comprises the steps of respectively processing original distribution data and rebalancing distribution data through a sharing aggregator to generate corresponding first image embedded representation and second image embedded representation, respectively inputting the first image embedded representation and the second image embedded representation into a first expert network and a second expert network to be processed to obtain a first prediction result and a second prediction result, determining consistency constraint loss through a consistency constraint unit based on the first prediction result and the second prediction result, and carrying out parameter optimization on an initial pathological image analysis model based on the consistency constraint loss to obtain a target pathological image analysis model. Compared with the prior art, the accuracy of the model on tail pathological category identification is improved.","assignee":"Shenzhen Shengqiang Technology Co ltd","inventors":["朱良慧","凌希通","李肖肖","陈玉玲","梁焯斌","黄强","申志远"],"publication_date":"2025-10-31","filing_date":"2025-09-29","priority_date":"2025-09-29","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30004"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120877071A/en"},{"publication_number":"CN120873992A","title":"Neural network-based rock-soil body deformation identification method and device and electronic equipment","abstract":"本发明提供了一种基于神经网络的岩土体变形识别方法、装置及电子设备，涉及人工智能技术领域，通过预先构建多源岩土融合张量，实现物理一致性和数据真实性的双重保障，并针对岩土体的监测数据对应的环境因子数据进行对应多源岩土融合张量匹配，实现具备物理力学一致性与现场数据真实性的特征识别。且，通过提取数据的分支特征、应变敏感特征和多尺度时空特征，形成覆盖力学本质、实时响应、时空演化的完整特征体系，对岩土体变形本质全面捕捉。并基于自注意力机制对上述特征协同分类，能够在适应不同地质条件的变形识别场景下精准确定变形模式，实现兼具物理一致性与数据驱动优势的岩土体变形分析。 This invention provides a method, device, and electronic device for identifying soil and rock deformation based on neural networks, belonging to the field of artificial intelligence technology. By pre-constructing a multi-source soil-rock fusion tensor, it achieves dual assurance of physical consistency and data authenticity. Furthermore, it performs corresponding multi-source soil-rock fusion tensor matching on environmental factor data corresponding to the monitoring data of the soil and rock mass, realizing feature recognition with both physical and mechanical consistency and on-site data authenticity. Moreover, by extracting branch features, strain-sensitive features, and multi-scale spatiotemporal features from the data, it forms a complete feature system covering the mechanical essence, real-time response, and spatiotemporal evolution, comprehensively capturing the essence of soil and rock deformation. Based on a self-attention mechanism, it collaboratively classifies the above features, accurately determining deformation patterns in deformation identification scenarios adapted to different geological conditions, achieving soil and rock deformation analysis with both physical consistency and data-driven advantages.","assignee":"Shandong Zhengyuan Construction Engineering Co ltd","inventors":["李亮亮","柴霞","高云忠","崔利新","徐强","王晓峰","肖敬瑞","荆瑞卿","王震"],"publication_date":"2025-10-31","filing_date":"2025-09-29","priority_date":"2025-09-29","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/251","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120873992A/en"},{"publication_number":"CN120873395A","title":"Regulation and control data optimization method and system based on living body implanted chip","abstract":"本申请公开了一种基于活体植入式芯片的调控数据优化方法及系统，属于数据处理技术领域。本申请利用植入式芯片持续收集活体内的生理数据，并基于优化激活特征和优化停止特征生成多个调控数据优化方案的关系图谱，使得优化策略的选择和执行更加科学、合理，避免了优化过程中的潜在影响因素和失误导致的数据波动。最终生成了以历史监测信息为优化基准，涉及各个时空域调控任务的目标优化指引信息，并将这一高度定制化的数据优化方案下发到活体植入式芯片，实现数据的精确预处理和优化，显著提升了数据的质量和可用性。 This application discloses a method and system for optimizing regulatory data based on an in vivo implantable chip, belonging to the field of data processing technology. This application utilizes an implantable chip to continuously collect physiological data in vivo and generates a relationship graph of multiple regulatory data optimization schemes based on optimized activation and cessation features. This makes the selection and execution of optimization strategies more scientific and reasonable, avoiding data fluctuations caused by potential influencing factors and errors during the optimization process. Ultimately, it generates target optimization guidance information covering various spatiotemporal regulatory tasks based on historical monitoring information, and distributes this highly customized data optimization scheme to the in vivo implantable chip, achieving precise data preprocessing and optimization, significantly improving data quality and usability.","assignee":"Ningbo Xinlian Xin Medical Technology Co ltd","inventors":["马骏"],"publication_date":"2025-10-31","filing_date":"2025-09-29","priority_date":"2025-09-29","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G16","G16H","G16H10/00","G","G16","G16H","G16H50/00","G16H50/70"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120873395A/en"},{"publication_number":"CN120869166A","title":"Multi-source data-based polar sea ice area route dynamic planning method","abstract":"本发明公开了一种基于多源数据的极地海域冰区航线动态规划方法，包括以下步骤：获取目标航道海域海冰、海流和风场历史数据，进行预处理；将预处理结果输入构建好的海冰预报模型中，得到规划日的海冰密集度和海冰厚度分布预报结果；将预报结果结合海流、风场预报数据，利用预先建立的随机森林航速推理模型，得到目标航道海域规划日航速空间分布数据；通过改进的A*算法，进行最小时间成本的航线规划，得到目标航道海域航线的动态规划结果。本发明涉及船舶海冰区航行技术领域，通过随机森林建模，拟合航速与海冰、海流、风场之间的关系，根据该拟合关系推理出极地海域航速背景场，进行最小时间成本的动态航线规划。 This invention discloses a dynamic route planning method for polar sea ice areas based on multi-source data, comprising the following steps: acquiring historical data on sea ice, ocean currents, and wind fields in the target route area and performing preprocessing; inputting the preprocessing results into a pre-constructed sea ice forecasting model to obtain the forecast results of sea ice concentration and sea ice thickness distribution for the planning day; combining the forecast results with ocean current and wind field forecast data, and using a pre-established random forest speed inference model to obtain the spatial distribution data of speed for the planning day in the target route area; and performing route planning with minimum time cost using an improved A* algorithm to obtain the dynamic planning result of the route in the target route area. This invention relates to the field of ship navigation technology in sea ice areas. By using random forest modeling to fit the relationship between speed and sea ice, ocean currents, and wind fields, the background speed field in polar sea areas is inferred based on this fitted relationship, and dynamic route planning with minimum time cost is performed.","assignee":"NATIONAL SATELLITE OCEAN APPLICATION SERVICE","inventors":["石立坚","王漫漫","邹斌","曾韬","路晓庆"],"publication_date":"2025-10-31","filing_date":"2025-09-29","priority_date":"2025-09-29","cpc_codes":["G","G01","G01C","G01C21/00","G01C21/20","G01C21/203","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/24323","G","G06","G06F","G06F18/00","G06F18/20","G06F18/27","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06Q","G06Q10/00","G06Q10/04","G06Q10/047","Y","Y02","Y02A","Y02A90/00","Y02A90/10"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120869166A/en"},{"publication_number":"CN120873153A","title":"Input text processing method, device, equipment, storage medium and program product","abstract":"The application provides an input text processing method, an input text processing device, input text processing equipment, a storage medium and a program product, and relates to the technical field of artificial intelligence (ARTIFICIAL INTELLIGENCE, AI). The method comprises the steps of extracting a reference named entity based on an input text processed by natural language, inquiring a reference text sentence associated with the reference named entity in a knowledge base, wherein the knowledge base comprises an association relation between the named entity and the text sentence, the text sentence is used for describing associated related knowledge of the named entity, and executing the natural language processing on the input text and the reference text sentence to obtain a natural language processing result of the input text. The scheme can improve the accuracy of executing natural language processing on the input text.","assignee":"Tencent Technology (Shenzhen) Co Ltd","inventors":["张子恒","林镇溪"],"publication_date":"2025-10-31","filing_date":"2024-04-28","priority_date":"2024-04-28","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/3332","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/30","G06F16/36","G06F16/367","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/289","G06F40/295","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G16","G16H","G16H50/00","G16H50/20"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120873153A/en"},{"publication_number":"CN115605877B","title":"Method, system and program product for a logical neural network","abstract":"A system for configuring and using a logical neural network comprising a graph syntax tree of formulas in a represented knowledge base connected to each other via nodes representing each proposition. There is one neuron for each logical connection that occurs in each formula, and furthermore, one neuron for each unique proposition that occurs in any formula. All neurons return pairs of values representing the upper and lower limits of the true values of their corresponding sub-formulas and propositions. Neurons corresponding to logical connections accept as inputs the outputs of neurons corresponding to their operands and have an activation function configured to match the connected truth function. Neurons corresponding to a proposition accept as input the output of neurons established as proof of the limits of the true value of the proposition, and have activation functions configured to aggregate the most tightly such limits. Bi-directional inference allows each occurrence of each proposition in each formula to be used as a potential proof.","assignee":"International Business Machines Corp","inventors":["R·里杰尔","F·卢斯","I·Y·阿克哈尔瓦亚","N·A·卡恩","N·马孔多","F·巴拉霍纳","A·格雷"],"publication_date":"2025-10-31","filing_date":"2021-04-13","priority_date":"2020-05-13","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/29","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06N","G06N5/00","G06N5/04","G06N5/042","G","G06","G06N","G06N5/00","G06N5/04","G06N5/046"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN115605877B/en"},{"publication_number":"CN120875676A","title":"Operation decision intelligent analysis method and system based on cross-domain data fusion","abstract":"本发明涉及数据分析技术领域，具体为一种基于跨域数据融合的运营决策智能分析方法及系统。首先，基于企业多领域本体知识库，通过基于语义映射的多源异构数据动态融合算法对来自不同业务系统的异构数据进行实体识别和关系映射，建立统一的数据模型；然后，采用融合因果推理和深度学习的混合智能决策引擎对该模型进行分析处理；接着，利用混合智能决策引擎的分析结果，通过因果关系发现算法构建业务变量间的多层次因果关系网络，并使用基于强化学习的自适应业务场景分析模型，根据业务环境变化动态调整分析策略，通过多目标优化算法生成帕累托最优决策方案集，输出运营决策建议；本发明提高企业运营决策智能分析的全面性与准确性。 This invention relates to the field of data analysis technology, specifically to an intelligent analysis method and system for operational decision-making based on cross-domain data fusion. First, based on an enterprise's multi-domain ontology knowledge base, a dynamic fusion algorithm based on semantic mapping is used to perform entity identification and relationship mapping on heterogeneous data from different business systems, establishing a unified data model. Then, a hybrid intelligent decision engine integrating causal reasoning and deep learning is used to analyze and process this model. Next, utilizing the analysis results from the hybrid intelligent decision engine, a multi-level causal relationship network among business variables is constructed through a causal relationship discovery algorithm. An adaptive business scenario analysis model based on reinforcement learning is then used to dynamically adjust the analysis strategy according to changes in the business environment. A Pareto optimal decision set is generated through a multi-objective optimization algorithm, outputting operational decision recommendations. This invention improves the comprehensiveness and accuracy of intelligent analysis for enterprise operational decisions.","assignee":"Beijing Shengbi Technology Co ltd","inventors":["李蜀毅","刘志锋","王楠楠"],"publication_date":"2025-10-31","filing_date":"2025-07-29","priority_date":"2025-07-29","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06393","G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/259","G","G06","G06F","G06F18/00","G06F18/20","G06F18/27","G","G06","G06F","G06F18/00","G06F18/20","G06F18/29","G06F18/295","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/12","G06N3/126","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/103","G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/105","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G","G06","G06Q","G06Q40/00","G06Q40/12","G06Q40/125","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120875676A/en"},{"publication_number":"CN120873974A","title":"Cross-modal knowledge graph driven wafer processing quality backtracking evaluation method","abstract":"The invention relates to the field of semiconductor manufacturing and discloses a cross-modal knowledge graph driven wafer processing quality backtracking evaluation method which comprises the following steps of obtaining and preprocessing time sequence process data and text log data, modeling the time sequence data by adopting a long-short-period memory network to predict defect probability, simultaneously extracting entities and relations from texts by adopting a BERT model to respectively construct process, defect and equipment knowledge graphs, fusing alignment cross-modal features of a generated countermeasure network into a unified cross-modal knowledge graph, carrying out probability backtracking reasoning of defect causes based on the unified graph, generating an active process parameter correction strategy, and outputting comprehensive quality scores by adopting a self-weighted dynamic evaluation model. The invention can realize intelligent closed loop from passive diagnosis to active optimization, remarkably improves the accuracy and efficiency of defect positioning, and provides a dynamic and comprehensive quality assessment means.","assignee":"Shenzhen University","inventors":["赵春洋","金世强","谢建龙"],"publication_date":"2025-10-31","filing_date":"2025-08-15","priority_date":"2025-08-15","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/254","G06F18/256","G","G06","G06F","G06F18/00","G06F18/20","G06F18/28","G","G06","G06F","G06F18/00","G06F18/20","G06F18/29","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06N","G06N7/00","G06N7/01","Y","Y02","Y02P","Y02P90/00","Y02P90/30"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120873974A/en"},{"publication_number":"KR20250156000A","title":"Multimodal artificial intelligence model training device for pet behavior analysis, and operating method thereof","abstract":"본 개시에 의하면, 행동 데이터 세트를 인코딩하고, 제1 및 제2 인코딩된 데이터를 출력하는 인코더, 및 인스트럭션 및 제2 인코딩된 데이터를 기반으로 생성된 임베딩 벡터, 및 제1 인코딩된 데이터를 기초로 멀티-헤드 어텐션을 계산하고, 멀티-헤드 어텐션의 어텐션 스코어들을 기초로 반려 동물을 해석한 텍스트를 나타내는 해석 데이터를 출력으로 예측하는 대규모 언어 모델 디코더를 포함하는 전자 장치 및 그 동작 방법을 제공한다. According to the present disclosure, an electronic device and a method of operating the same are provided, including an encoder that encodes a behavioral data set and outputs first and second encoded data, an embedding vector generated based on instructions and the second encoded data, and a large-scale language model decoder that calculates multi-head attention based on the first encoded data and predicts interpretation data representing a text interpreted about a companion animal based on attention scores of the multi-head attention.","assignee":"정소영","inventors":["정소영"],"publication_date":"2025-10-31","filing_date":"2024-12-23","priority_date":"2024-04-24","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","A","A01","A01K","A01K11/00","A01K11/006","A01K11/008","A","A01","A01K","A01K29/00","A01K29/005","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V40/00","G06V40/20","G","G10","G10L","G10L15/00","G10L15/06","G10L15/063","G","G10","G10L","G10L25/00","G10L25/27","G10L25/30","G","G10","G10L","G10L25/00","G10L25/48","G10L25/51","G10L25/63","A","A01","A01K","A01K27/00","A01K27/001","A","A01","A01K","A01K27/00","A01K27/002"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250156000A/en"},{"publication_number":"CN120876183A","title":"Knowledge tracking method based on mixed convolution","abstract":"本发明涉及知识追踪技术领域，特别是涉及基于混合卷积的知识追踪方法，包括：获取学习者与教育内容的交互信息；将交互信息输入预设的知识追踪模型中，预测目标题目的答题正确概率，其中，所述知识追踪模型用于基于所述交互信息构建输入序列，利用多尺度因果卷积提取序列中不同时间范围的行为特征，再结合带有距离惩罚项的注意力机制进行建模，最终结合目标题目的答题正确概率预测完成知识追踪。本发明结合多尺度因果卷积与距离衰减注意力机制进行知识追踪，提升对学习者认知状态的建模能力。 This invention relates to the field of knowledge tracing technology, and in particular to a knowledge tracing method based on hybrid convolution, comprising: acquiring interaction information between learners and educational content; inputting the interaction information into a preset knowledge tracing model to predict the probability of correctly answering a target question; wherein the knowledge tracing model is used to construct an input sequence based on the interaction information, extract behavioral features of different time ranges in the sequence using multi-scale causal convolution, and then combine it with an attention mechanism with a distance penalty term for modeling, and finally combine the prediction of the probability of correctly answering the target question to complete knowledge tracing. This invention combines multi-scale causal convolution and a distance decay attention mechanism for knowledge tracing, improving the ability to model learners' cognitive states.","assignee":"Jinan University","inventors":["刘子韬","侯明良"],"publication_date":"2025-10-31","filing_date":"2025-09-26","priority_date":"2025-09-26","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06Q","G06Q10/00","G06Q10/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120876183A/en"},{"publication_number":"CN120874515A","title":"Parameter exploration method","abstract":"提供一种半导体元件的参数候补。参数抽出部被供应测量数据作为数据组而抽出模型参数。电路模拟器被供应第一网表，利用第一网表及模型参数进行模拟，输出第一输出结果。分类模型学习模型参数及第一输出结果而对模型参数进行分类。电路模拟器被供应第二网表及模型参数。神经网络被供应要调整的变数，输出动作值函数而更新变数。电路模拟器利用第二网表及模型参数进行模拟，在输出的第二输出结果不满足条件时更新神经网络的权重系数而在满足条件时将变数判定为最佳候选。 A parameter candidate for a semiconductor device is provided. A parameter extraction unit is supplied with measurement data as a data set to extract model parameters. A circuit simulator is supplied with a first netlist, performs simulation using the first netlist and the model parameters, and outputs a first output result. A classification model learns the model parameters and the first output result to classify the model parameters. The circuit simulator is supplied with a second netlist and model parameters. A neural network is supplied with variables to be adjusted, outputs an action-value function to update the variables. The circuit simulator performs simulation using the second netlist and the model parameters, updates the weight coefficients of the neural network when the second output result does not meet the conditions, and determines the variable as the best candidate when the conditions are met.","assignee":"Semiconductor Energy Laboratory Co Ltd","inventors":["小国哲平","长多刚","福留贵浩"],"publication_date":"2025-10-31","filing_date":"2020-02-04","priority_date":"2019-02-15","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F30/00","G06F30/30","G06F30/36","G","G06","G06F","G06F30/00","G06F30/30","G06F30/36","G06F30/367","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120874515A/en"},{"publication_number":"CN120728750B","title":"A Virtual Power Plant Response Optimization Scheduling System and Method Based on Reinforcement Learning","abstract":"The invention discloses a virtual power plant response optimization scheduling system and method based on reinforcement learning, and relates to the technical field of virtual power plant intelligent scheduling. The system comprises an environment modeling module, an intelligent agent module, a multi-agent coordination module and a self-adaptive optimization module, wherein the environment modeling module, the intelligent agent module, the multi-agent coordination module and the self-adaptive optimization module are respectively used for constructing a multidimensional state space and a layered action space, generating and optimizing action strategies based on an Actor-Critic network, executing scheduling instructions through a layered multi-agent structure and realizing conflict consensus, and dynamically adapting to state space changes by combining incremental learning and meta-learning mechanisms. The system and the method have the advantages of fine state modeling, high action response efficiency, self-adaption of strategy updating, cooperative stability of the intelligent agent and the like, and can keep the continuity, stability and optimality of the scheduling strategy in an operating environment with cooperative participation of multi-source heterogeneous power resources in scheduling, frequent change of market rules and severe load fluctuation.","assignee":"Nanjing Nanzi Huadun Digital Technology Co ltd","inventors":["周新亚","戚明喆","陈震","黄保乐","冷程浩","赵竟","高波","李安多"],"publication_date":"2025-10-31","filing_date":"2025-08-21","priority_date":"2025-08-21","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/06","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","H","H02","H02J","H02J3/00","H02J3/28","H","H02","H02J","H02J3/00","H02J3/38","H02J3/381","H","H02","H02J","H02J3/00","H02J3/38","H02J3/46","H02J3/466","H","H02","H02J","H02J2101/00","H02J2101/20","H02J2101/22","H02J2101/24","H","H02","H02J","H02J2101/00","H02J2101/20","H02J2101/28","H","H02","H02J","H02J2103/00","H02J2103/30","Y","Y04","Y04S","Y04S10/00","Y04S10/50"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120728750B/en"},{"publication_number":"CN120872714A","title":"AI chip test parameter self-adaptive optimization method based on deep learning","abstract":"本发明涉及深度学习技术领域，具体为一种基于深度学习的AI芯片测试参数自适应优化方法，包括采集AI芯片的历史测试数据，基于所述历史测试数据计算不同失效模式之间的相关性强度；根据边权值识别失效耦合矩阵，将预设的静态检测参数约束边界转换为随失效检测状态变化的动态约束空间；构建多层级优化架构，上层执行失效类型相关性分析并生成约束传播信息，中层基于约束传播信息对参数簇进行优化，下层对单个检测参数进行调节，输出参数优化结果；建立神经网络映射模型，获取检测参数与失效类型之间的非线性映射关系；基于物理状态参数，调整非线性映射关系，更新动态约束空间，并重新执行参数优化，输出参数优化结果，输出最优测试参数组合。 This invention relates to the field of deep learning technology, specifically to an adaptive optimization method for AI chip test parameters based on deep learning. The method includes: collecting historical test data of the AI chip; calculating the correlation strength between different failure modes based on the historical test data; identifying the failure coupling matrix according to edge weights; converting the preset static detection parameter constraint boundary into a dynamic constraint space that changes with the failure detection state; constructing a multi-level optimization architecture, where the upper layer performs failure type correlation analysis and generates constraint propagation information, the middle layer optimizes parameter clusters based on constraint propagation information, and the lower layer adjusts individual detection parameters, outputting the parameter optimization results; establishing a neural network mapping model to obtain the nonlinear mapping relationship between detection parameters and failure types; adjusting the nonlinear mapping relationship based on physical state parameters, updating the dynamic constraint space, re-executing parameter optimization, outputting the parameter optimization results, and outputting the optimal test parameter combination.","assignee":"Jiangsu Haina Electronic Technology Co ltd","inventors":["陶雪峰","郑超","代彬","段宏宇"],"publication_date":"2025-10-31","filing_date":"2025-09-26","priority_date":"2025-09-26","cpc_codes":["G","G06","G06F","G06F11/00","G06F11/22","G06F11/2273","G","G06","G06F","G06F11/00","G06F11/22","G06F11/2205","G","G06","G06F","G06F18/00","G06F18/20","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2431","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/254","G06F18/256","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120872714A/en"},{"publication_number":"CN113762047B","title":"Method and device for predicting residual available time length of image acquisition device and electronic equipment","abstract":"本申请实施例提供了一种图像采集装置的剩余可用时长预测方法、装置及电子设备，涉及设备检测技术领域。该方法包括：获取目标图像采集装置在至少一个维度的运行状态信息；将每个维度的运行状态信息输入通过人工智能技术训练的可用时长预测模型，获得可用时长预测模型输出的目标图像采集装置在相应维度的影响因子；根据目标图像采集装置在每个维度的影响因子，获得目标图像采集装置的剩余可用时长。本申请实施例能够获得图像采集装置的高准确度的剩余可用时长，并且训练可用时长预测模型的效率也较相关技术更高。 This application provides a method, apparatus, and electronic device for predicting the remaining available time of an image acquisition device, relating to the field of device detection technology. The method includes: acquiring operational status information of the target image acquisition device in at least one dimension; inputting the operational status information of each dimension into an available time prediction model trained using artificial intelligence technology to obtain the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model; and obtaining the remaining available time of the target image acquisition device based on the influence factor of the target image acquisition device in each dimension. This application embodiment can obtain the remaining available time of the image acquisition device with high accuracy, and the efficiency of training the available time prediction model is also higher than related technologies.","assignee":"Tencent Technology (Shenzhen) Co Ltd","inventors":["赵伟","郭润增","王少鸣","洪哲鸣","王军","彭旭康"],"publication_date":"2025-10-31","filing_date":"2021-05-07","priority_date":"2021-05-07","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q20/00","G06Q20/38","G06Q20/40","G06Q20/401","G06Q20/4014","G06Q20/40145"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN113762047B/en"},{"publication_number":"CN116168252B","title":"Image emotion recognition method based on abstract relation scene graph","abstract":"本发明公开了一种基于抽象关系场景图的图像情感识别方法，包括以下步骤：构建新的对象和属性检测器并提取图像中对象特征及其对应的属性特征；通过对象的特征推理对象间的亲密度和抽象关系特征并构建抽象关系场景图；构建图神经网络推理抽象关系场景图使各对象的特征包含情感因素；构建场景特征提取器提取图像的场景特征并设计渐进式注意力机制融合多个对象特征得到一个整体对象特征；拼接图像的场景特征和整体对象特征送入情感分类器得到图像情感类别。相较于现有技术，本发明利用对象间的关系以及对象与场景间的交互作用识别图像情感，有效缩小了低级视觉和高级情感之间的鸿沟，并提升了图像情感识别的分类准确率。 This invention discloses an image emotion recognition method based on an abstract relationship scene graph, comprising the following steps: constructing a new object and attribute detector and extracting object features and their corresponding attribute features from the image; inferring the intimacy and abstract relationship features between objects through object features and constructing an abstract relationship scene graph; constructing a graph neural network to infer the abstract relationship scene graph so that the features of each object include emotional factors; constructing a scene feature extractor to extract the scene features of the image and designing a progressive attention mechanism to fuse multiple object features to obtain a holistic object feature; stitching together the scene features and the holistic object feature of the image and feeding them into an emotion classifier to obtain the image emotion category. Compared with existing technologies, this invention utilizes the relationships between objects and the interaction between objects and the scene to identify image emotions, effectively narrowing the gap between low-level vision and high-level emotions, and improving the classification accuracy of image emotion recognition.","assignee":"Guilin University of Electronic Technology","inventors":["文益民","康博"],"publication_date":"2025-10-31","filing_date":"2023-03-07","priority_date":"2023-03-07","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","Y","Y02","Y02D","Y02D10/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN116168252B/en"},{"publication_number":"KR20250155492A","title":"Automated system and method for contracts and payments for energy saving performance rewards","abstract":"본 발명은 다중 에너지 절감 실적의 실시간 수집과 검증, 계약 조건에 기반한 보상금 산출, 자동 결제, 그리고 스마트컨트랙트를 활용한 데이터 신뢰성 확보 및 법ㆍ조례 연계까지 포함하는 성과보상형 계약ㆍ결제 자동화 시스템 및 방법을 제공한다. 이를 통해 공공기관 및 일반 기업에서 투명하고 효율적인 에너지 절감 계약과 결제 문화 확산에 기여하며, 산업 경쟁력 및 지속가능 발전에 이바지한다. The present invention provides a performance-based contract and payment automation system and method that includes real-time collection and verification of multiple energy saving performance, calculation of compensation based on contract terms, automatic payment, and securing data reliability and linking with laws and regulations using smart contracts. Through this, we contribute to the spread of transparent and efficient energy-saving contracts and payment culture in public institutions and private companies, and contribute to industrial competitiveness and sustainable development.","assignee":"구교선","inventors":["구교선"],"publication_date":"2025-10-30","filing_date":"2025-10-13","priority_date":"2025-10-13","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/06","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/10","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26","G","G16","G16Y","G16Y10/00","G16Y10/35","G","G16","G16Y","G16Y40/00","G16Y40/10","Y","Y04","Y04S","Y04S50/00","Y04S50/12","Y","Y04","Y04S","Y04S50/00","Y04S50/16"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250155492A/en"},{"publication_number":"AU2025248733A1","title":"Missed-Bolus Dose Detection And Related Systems, Methods And Devices","abstract":"Disclosed embodiments relate, generally, to retrospective missed-bolus detection. Some embodiments relate to systems, methods, and devices for performing retrospective missed-bolus detection by processing insulin therapy data. Some embodiments relate, generally, to systems, methods and devices for training missed-bolus classifiers using machine learning techniques to perform retrospective missed-bolus detection. Some embodiments relate, generally, to systems, methods, and devices for obtaining training data and test data that may be used to train missed-bolus classifiers to perform retrospective missed-bolus detection. (FIG. 1)","assignee":"Bigfoot Biomedical Inc","inventors":["Alexandra Elena CONSTANTIN","Zahra EGHTESADI"],"publication_date":"2025-10-30","filing_date":"2025-10-10","priority_date":"2019-06-10","cpc_codes":["G","G16","G16H","G16H20/00","G16H20/10","G","G16","G16H","G16H20/00","G16H20/10","G16H20/17","G","G06","G06N","G06N20/00","G","G16","G16H","G16H20/00","G16H20/60","G","G16","G16H","G16H50/00","G16H50/70"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025248733A1/en"},{"publication_number":"AU2025248637A1","title":"Intelligent recommendations based on multimodal inputs","abstract":"A computing system includes a memory device and a processor structured to: receive, from a user device, a media input; extract, using a machine-learning model, at least one feature of the media input; identify, using the machine-learning model, at least one intent; determine, using the machine-learning model, at least one action policy; generate a user interface comprising the media input, the at least one extracted feature of the media input, and the at least one action policy; provide the user interface to the user device; receive, via an input to the user interface, an indication of a selection of the at least one action policy displayed on the user interface; generate a second user interface comprising a plurality of options associated with the selected at least one action policy; and provide the second user interface to the user device.","assignee":"Expedia Inc","inventors":["Landon CHAMBERS","Raisa Meneses GUZMAN","Travers HUMBLE","Anush Kumar","Nehemias LUNA","Marc Silbey","Justin SIMONELLI","Phalguna YADLAPATI"],"publication_date":"2025-10-30","filing_date":"2025-10-07","priority_date":"2023-06-21","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G06F16/9535","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0631","G","G06","G06N","G06N20/00"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025248637A1/en"},{"publication_number":"US20250334942A1","title":"AI-Based Energy Edge Platform, Systems, and Methods Having Automated and Coordinated Governance of Resource Sets","abstract":"An AI-based platform for enabling intelligent orchestration and management of power and energy is disclosed. The platform includes a system configured to perform automated and coordinated governance of a set of energy entities that are operationally coupled within an energy grid and a set of distributed edge energy resources. At least one of the distributed edge energy resources is operationally independent of the energy grid.","assignee":"Strong Force Ee Portfolio 2022 LLC","inventors":["Charles Howard Cella","Andrew Cardno"],"publication_date":"2025-10-30","filing_date":"2025-05-09","priority_date":"2021-11-23","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/06","G","G01","G01R","G01R21/00","G01R21/133","G","G05","G05B","G05B13/00","G05B13/02","G05B13/0265","G","G05","G05B","G05B13/00","G05B13/02","G05B13/04","G","G05","G05B","G05B13/00","G05B13/02","G05B13/04","G05B13/042","G","G05","G05B","G05B19/00","G05B19/02","G05B19/04","G05B19/042","G","G06","G06F","G06F1/00","G06F1/26","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/245","G06F18/2453","G","G06","G06N","G06N10/00","G","G06","G06N","G06N10/00","G06N10/80","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/04","G06N5/043","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/067","G","G06","G06Q","G06Q30/00","G06Q30/018","G","G06","G06Q","G06Q50/00","G06Q50/02","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26","G","G06","G06Q","G06Q99/00","G","G06","G06V","G06V10/00"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250334942A1/en"},{"publication_number":"US20250335326A1","title":"Systems and methods for determining a user specific mission operational performance metric, using machine-learning processes","abstract":"Aspects relate to system and methods for determining a user specific mission operational performance, using machine-learning processes. An exemplary system includes a computing device configured to perform operations including receiving user-input structured data from at least a user device, receiving observed structured data related to the user and a mission performance metric, inputting the user-input structured data and the observed structured data to a machine-learning model, generating a user performance metric as a function of the machine-learning model, receiving a deterministic mission operational performance metric, disaggregating a deterministic user performance metric as a function of the deterministic mission operation performance metric and the mission performance metric, inputting training data to a machine-learning algorithm, where the training data includes the user-input structured data and the observed structured data correlated to the deterministic user performance metric, and training the machine-learning model as a function of the machine-learning algorithm and the training data.","assignee":"Gmeci LLC","inventors":["Bradford R. EVERMAN","Brian Scott Bradke"],"publication_date":"2025-10-30","filing_date":"2025-03-11","priority_date":"2021-05-28","cpc_codes":["G","G06","G06F","G06F11/00","G06F11/30","G06F11/34","G06F11/3409","G06F11/3428","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G06F18/2155","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G06F18/24155","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2431","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N5/00","G06N5/01"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250335326A1/en"},{"publication_number":"KR102878791B1","title":"Method, device and system for providing local authentication automation platform service and matching of supplier and local sales outlet for advancing to reverse direction of k-beauty product based on artificial intelligence model","abstract":"일실시예에 따르면, 장치에 의해 수행되는, 인공지능 모델 기반 K-뷰티 제품의 역직구 진출을 위한 공급처 및 현지 판매처의 매칭과 현지 인증 자동화 플랫폼 서비스 제공 방법에 있어서, 제1 국가의 역직구 진출이 필요한 K-뷰티 제품으로 제1 제품이 선정된 경우, 상기 제1 제품의 카테고리, 타겟 고객층 및 효능을 확인한 결과를 기반으로, 상기 제1 제품의 특징 정보를 생성하는 단계; 상기 제1 제품의 카테고리가 제1 카테고리인 것으로 확인되면, 국내에서 상기 제1 카테고리의 제품에 대한 공급이 가능한 것으로 확인된 공급처를 후보 공급처로 분류하는 단계; 제1 공급처가 상기 후보 공급처로 분류된 경우, 상기 제1 공급처의 제품 공급 내역을 기반으로, 상기 제1 공급처의 활동 정보를 생성하는 단계; 상기 제1 제품의 특징 정보 및 상기 제1 공급처의 활동 정보를 매칭하여, 제1 매칭 정보를 생성하는 단계; 상기 제1 매칭 정보를 인코딩 하여 제1 입력 신호를 생성하는 단계; 제품의 특징과 공급처의 활동 상태를 고려하여 제품과 공급처 간의 매칭 점수를 분석하도록 학습된 제1 인공지능 모델에 상기 제1 입력 신호를 입력하는 단계; 상기 제1 입력 신호를 통해 상기 제1 제품과 상기 제1 공급처 간의 매칭 점수가 제1 점수로 산출되면, 상기 제1 인공지능 모델로부터 상기 제1 점수를 나타내는 제1 출력 신호를 획득하는 단계; 상기 제1 출력 신호를 기초로, 상기 제1 제품과 상기 제1 공급처 간의 매칭 점수를 상기 제1 점수로 설정하는 단계; 상기 제1 제품과 상기 후보 공급처로 분류된 공급처들 간의 매칭 점수가 각각 설정된 경우, 각각 설정된 매칭 점수를 비교한 결과, 상기 후보 공급처로 분류된 공급처들 중에서 상기 제1 공급처의 매칭 점수가 가장 높은 것으로 확인되면, 상기 제1 공급처를 상기 제1 제품의 공급처로 선정하는 단계; 상기 제1 국가에서 상기 제1 카테고리의 제품에 대한 판매가 가능한 것으로 확인된 판매처를 후보 판매처로 분류하는 단계; 제1 판매처가 상기 후보 판매처로 분류된 경우, 상기 제1 판매처의 제품 판매 내역을 기반으로, 상기 제1 판매처의 활동 정보를 생성하는 단계; 상기 제1 제품의 특징 정보 및 상기 제1 판매처의 활동 정보를 매칭하여, 제2 매칭 정보를 생성하는 단계; 상기 제2 매칭 정보를 인코딩 하여 제2 입력 신호를 생성하는 단계; 제품의 특징과 판매처의 활동 상태를 고려하여 제품과 판매처 간의 매칭 점수를 분석하도록 학습된 제2 인공지능 모델에 상기 제2 입력 신호를 입력하는 단계; 상기 제2 입력 신호를 통해 상기 제1 제품과 상기 제1 판매처 간의 매칭 점수가 제2 점수로 산출되면, 상기 제2 인공지능 모델로부터 상기 제2 점수를 나타내는 제2 출력 신호를 획득하는 단계; 상기 제2 출력 신호를 기초로, 상기 제1 제품과 상기 제1 판매처 간의 매칭 점수를 상기 제2 점수로 설정하는 단계; 상기 제1 제품과 상기 후보 판매처로 분류된 판매처들 간의 매칭 점수가 각각 설정된 경우, 각각 설정된 매칭 점수를 비교한 결과, 상기 후보 판매처로 분류된 판매처들 중에서 상기 제1 판매처의 매칭 점수가 가장 높은 것으로 확인되면, 상기 제1 판매처를 상기 제1 제품의 판매처로 선정하는 단계; 및 상기 제1 공급처와 상기 제1 판매처를 상기 제1 제품의 역직구 진출을 위한 공급처와 현지 판매처로 매칭하는 단계를 포함하는, 인공지능 모델 기반 K-뷰티 제품의 역직구 진출을 위한 공급처 및 현지 판매처의 매칭과 현지 인증 자동화 플랫폼 서비스 제공 방법이 제공된다. According to one embodiment, a method for providing a platform service for matching suppliers and local sales outlets and for local authentication automation for K-beauty products for direct import from a first country, based on an artificial intelligence model, is provided, the method comprising: a step of generating characteristic information of the first product based on a result of confirming the category, target customer base, and efficacy of the first product when a first product is selected as a K-beauty product requiring direct import from a first country; a step of classifying suppliers that are confirmed to be able to supply products of the first category domestically as candidate suppliers when the category of the first product is confirmed to be the first category; a step of generating activity information of the first supplier based on a product supply history of the first supplier when the first supplier is classified as the candidate supplier; a step of matching the characteristic information of the first product and the activity information of the first supplier to generate first matching information; a step of encoding the first matching information to generate a first input signal; a step of inputting the first input signal into a first artificial intelligence model trained to analyze a matching score between the product and the supplier in consideration of the characteristics of the product and the activity status of the supplier; When a matching score between the first product and the first supplier is calculated as a first score through the first input signal, a step of obtaining a first output signal representing the first score from the first artificial intelligence model; a step of setting the matching score between the first product and the first supplier as the first score based on the first output signal; a step of selecting the first supplier as a supplier of the first product when the matching scores between the first product and the suppliers classified as candidate suppliers are each set and, as a result of comparing the set matching scores, it is confirmed that the matching score of the first supplier is the highest among the suppliers classified as candidate suppliers; a step of classifying a seller confirmed to be able to sell a product of the first category in the first country as a candidate seller; a step of generating activity information of the first seller based on the product sales history of the first seller when the first seller is classified as the candidate seller; a step of generating second matching information by matching the characteristic information of the first product and the activity information of the first seller; A step of encoding the second matching information to generate a second input signal; A step of inputting the second input signal to a second artificial intelligence model trained to analyze a matching score between a product and a seller by considering the characteristics of the product and the activity status of the seller; A step of obtaining a second output signal representing the second score from the second artificial intelligence model when the matching score between the first product and the first seller is calculated as a second score through the second input signal; A step of setting the matching score between the first product and the first seller as the second score based on the second output signal; A step of selecting the first seller as the seller of the first product when the matching scores between the first product and the sellers classified as candidate sellers are each set and, as a result of comparing the set matching scores, it is confirmed that the matching score of the first seller is the highest among the sellers classified as candidate sellers. A method for matching suppliers and local sellers for direct purchase of K-beauty products based on an artificial intelligence model and providing a local authentication automation platform service is provided, including a step of matching the first supplier and the first seller with suppliers and local sellers for direct purchase of the first product.","assignee":"주식회사 디아스포라","inventors":["박선민"],"publication_date":"2025-10-30","filing_date":"2025-02-28","priority_date":"2025-02-28","cpc_codes":["G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0613","G06Q30/0619","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q10/00","G06Q10/08","G06Q10/083","G06Q10/0831","G","G06","G06Q","G06Q30/00","G06Q30/018"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102878791B1/en"},{"publication_number":"DK202530631A1","title":"A method of determining an event of an industrial process at an industrial site and a system thereof","abstract":"The present invention relates to a method of determining at least one event of an industrial process at an industrial site, comprising acts of collecting sensor data representing conditions at/or related to the industrial site, inputting the collected data to a data processing apparatus, performing data analysis on the one or more of the inputted data using feature extraction for determine at least one event, logging a timestamp and a position of the at least one determined event, and synchronising the at least one event with the data inputted to the data processing apparatus to form a structured dataset, storing said structured dataset in a database. The time stamp is synchronized to an external clock operating a globally unique time system. The present invention also relates to a system configured to perform the method. The present invention further relates to a computer program configured to execute the method and a computer-readable medium configured to store the computer program.","assignee":"Claviate Aps","inventors":["Koops Kratmann Kasper","Aron Gudnason Daniel"],"publication_date":"2025-10-29","filing_date":"2025-10-13","priority_date":"2024-02-21","cpc_codes":["G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6245","G06F21/6254","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G","G06","G06V","G06V20/00","G06V20/40","G06V20/44","G","G06","G06V","G06V20/00","G06V20/50","G06V20/52","H","H04","H04L","H04L63/00","H04L63/04","H04L63/0428"],"country":"DK","kind":"application","source_url":"https://patents.google.com/patent/DK202530631A1/en"},{"publication_number":"KR20250154990A","title":"AI-based Multimedia-Integrated Official Document Auto-Generation and Blockchain Security System","abstract":"본 발명은 AI 기반 멀티미디어 통합 공식문서 자동생성 및 블록체인 보안 시스템에 관한 것으로, 음성, 텍스트, 이미지, 영상 등 다양한 멀티미디어 데이터를 AI가 자동 분석하여 정부 표준양식에 맞는 공식문서를 생성하고, 블록체인 기술로 문서의 위변조를 방지한다. 본 시스템은 멀티미디어 입력부(100), AI 분석 엔진(200), 문서 생성 엔진(300), 블록체인 보안부(400), 질의응답 시스템(500)을 포함하며, 문서 작성 시간을 95% 단축하고 위변조를 100% 방지한다. 또한 업로드된 문서에 대한 예상 질문을 자동 생성하고 답변서를 작성하여 질의응답 시간을 99.5% 단축한다. 이를 통해 정부 업무의 완전 자동화 및 디지털 전환을 실현한다. The present invention relates to an AI-based multimedia integrated official document automatic generation and blockchain security system. The system automatically analyzes various multimedia data, such as voice, text, images, and video, to generate official documents conforming to government standard formats, and prevents document forgery and falsification using blockchain technology. The system comprises a multimedia input unit (100), an AI analysis engine (200), a document generation engine (300), a blockchain security unit (400), and a question-and-answer system (500), reducing document creation time by 95% and preventing forgery and falsification by 100%. Furthermore, the system automatically generates expected questions for uploaded documents and prepares answers, reducing question-and-answer time by 99.5%. This realizes complete automation and digital transformation of government work.","assignee":"문지영","inventors":["문지영"],"publication_date":"2025-10-29","filing_date":"2025-10-11","priority_date":"2025-10-11","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/93","G","G06","G06F","G06F16/00","G06F16/20","G06F16/21","G06F16/219","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F16/00","G06F16/30","G06F16/34","G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9032","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9038","G","G06","G06F","G06F21/00","G06F21/60","G06F21/64","G","G06","G06F","G06F40/00","G06F40/10","G06F40/103","G","G06","G06F","G06F40/00","G06F40/10","G06F40/166","G06F40/174","G","G06","G06F","G06F40/00","G06F40/10","G06F40/166","G06F40/186","G","G06","G06F","G06F40/00","G06F40/10","G06F40/194","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26","G","G06","G06V","G06V20/00","G06V20/80","G","G06","G06V","G06V30/00","G06V30/40","G","G10","G10L","G10L15/00","G10L15/26","H","H04","H04L","H04L9/00","H04L9/06","H04L9/0643","H","H04","H04N","H04N5/00","H04N5/222","H04N5/262","H04N5/278"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250154990A/en"},{"publication_number":"KR20250154355A","title":"Apparatus and method for generating document images used in machine-learning of text detection and recognition","abstract":"본 발명은 학습 데이터의 생성 기술에 관한 것으로서, 상세하게는 이미지 내에 존재하는 텍스트를 검출 및 인식하는 모델을 학습할 때 학습 데이터로 사용되는 문서 이미지를 생성하기 위한 문서 이미지의 생성 장치 및 방법에 관한 것이다. 이를 위해, 본 발명에 따른 문서 이미지의 생성 방법은 컴퓨팅 장치에서 수행되는 문서 이미지의 생성 방법으로서, 정답 단어 박스 및 문자열 정보를 가진 문서 이미지를 입력받아 문서 이미지 내의 단어 박스의 위치를 검출하는 단계와, 상기 문서 이미지 내의 단어 박스에 있는 글자 영역을 배경 이미지로 생성하는 단계와, 상기 배경 이미지가 된 글자 영역에 임의의 폰트로 임의의 글자를 그려 새로운 문서 이미지를 생성하는 단계를 포함한다. The present invention relates to a technology for generating learning data, and more particularly, to a device and method for generating document images for generating document images used as learning data when training a model for detecting and recognizing text existing in an image. To this end, a method for generating a document image according to the present invention is a method for generating a document image performed in a computing device, comprising: a step of receiving a document image having a correct word box and string information and detecting the location of the word box within the document image; a step of generating a character area within the word box within the document image as a background image; and a step of drawing arbitrary characters in an arbitrary font in the character area that has become the background image to generate a new document image.","assignee":"주식회사 하나금융티아이","inventors":["문지영","여동훈","윤인용","김수현"],"publication_date":"2025-10-28","filing_date":"2025-10-22","priority_date":"2022-04-06","cpc_codes":["G","G06","G06V","G06V30/00","G06V30/10","G06V30/19","G06V30/191","G06V30/19147","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V30/00","G06V30/10","G06V30/20","G","G06","G06V","G06V30/00","G06V30/40","G06V30/41","G06V30/413"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250154355A/en"},{"publication_number":"KR20250154579A","title":"Method, system, and computer program for personalized recommendation based on topic of interest","abstract":"관심 주제 기반 개인화 추천을 위한 방법, 시스템, 및 컴퓨터 프로그램이 개시된다. 인터넷 상에서 사용되는 키워드에 대해 언어 모델(language model)을 이용하여 상기 키워드의 주제를 생성하여 상기 키워드와 상기 주제 간의 관계를 구축하고, 상기 키워드와 상기 주제 간의 관계를 이용하여 인터넷 상의 사용자의 활동에 대응되는 적어도 하나의 관심 주제를 선정하고, 상기 관심 주제를 기초로 상기 사용자에 대해 개인화된 추천 정보를 제공할 수 있다. A method, system, and computer program for personalized recommendation based on topic of interest are disclosed. The method comprises generating topics for keywords used on the Internet using a language model, establishing relationships between the keywords and the topics, selecting at least one topic of interest corresponding to a user's activity on the Internet using the relationship between the keywords and the topics, and providing personalized recommendation information to the user based on the topic of interest.","assignee":"네이버 주식회사","inventors":["김동현","김충기","양민철","황현동","유혜수","장예훈","이정태"],"publication_date":"2025-10-28","filing_date":"2025-10-21","priority_date":"2022-01-25","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G06F16/9535","G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G06F16/9024","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9032","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9035","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9038","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G06F16/9532","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G06F16/9538","G","G06","G06N","G06N20/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250154579A/en"},{"publication_number":"KR20250154578A","title":"Apparatus and method for customized stress care","abstract":"본 발명은 스트레스 측정 및 케어 기술에 관한 것으로서, 상세하게는 인체 동영상 및 스트레스 문항 검사결과 데이터를 분석해 스트레스 측정결과를 얻고 그에 따른 스트레스 치유 프로그램의 각 케어 과정을 진행하면서 스트레스를 반복 측정하여 스트레스 치유 피드백을 제공할 수 있는 사용자 맞춤형 스트레스 케어 장치 및 방법에 관한 것이다. 이를 위해, 본 발명에 따른 사용자 맞춤형 스트레스 케어 장치는 사용자에게 설문 항목을 제공하여 스트레스 및 내적 취약성 검사결과 데이터를 생성하는 스트레스 설문 검사부와, 상기 검사결과 데이터 및 인체 동영상을 입력 받아 회복 탄력성 및 스트레스 고위험군 예측결과인 스트레스 측정결과를 출력하는 스트레스 측정부와, 상기 스트레스 측정결과에 따른 스트레스 치유 프로그램을 실행하는 케어 프로그램 제공부를 포함한다. The present invention relates to stress measurement and care technology, and more particularly, to a user-customized stress care device and method capable of analyzing human body video and stress item test result data to obtain stress measurement results, repeatedly measuring stress while proceeding with each care process of a stress healing program according to the stress measurement results, and providing stress healing feedback. To this end, the user-customized stress care device according to the present invention includes a stress questionnaire test unit that provides a user with questionnaire items to generate stress and internal vulnerability test result data, a stress measurement unit that receives the test result data and human body video and outputs a stress measurement result that is a prediction result of resilience and a high-risk group for stress, and a care program providing unit that executes a stress healing program according to the stress measurement results.","assignee":"(주)포티파이","inventors":["문성준","노경진","성기영","문우리"],"publication_date":"2025-10-28","filing_date":"2025-10-14","priority_date":"2022-11-01","cpc_codes":["G","G16","G16H","G16H20/00","G16H20/70","A","A61","A61B","A61B5/00","A61B5/02","A61B5/024","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4884","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7275","A","A61","A61M","A61M21/00","G","G06","G06N","G06N20/00","G","G16","G16H","G16H10/00","G16H10/20","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H30/00","G16H30/20","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/70"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250154578A/en"},{"publication_number":"KR20250154323A","title":"Server for performing multi-stage parameter learning for efficient vocabulary expansion of multilingual llm and method for operation thereof","abstract":"다양한 실시예들에 따라서, 다국어 LLM을 실현하기 위하여 모델 파라미터에 대한 단계별 학습을 수행하는 서버는, 메모리, 통신 모듈, 및 프로세서를 포함하고, 상기 프로세서는, 미리 학습된 토크나이저(tokenizer)를 이용하여 미리 정해진 토큰 세트에 추가 토큰 세트를 병합한 학습용 토큰 세트를 생성하고, 상기 추가 토큰 세트에 대하여 서브워드 기반의 임베딩 초기화를 수행하고, 다국어 LLM의 트랜스포머 계층의 학습에 사용되는 복수의 제1 파라미터 그룹, 상기 미리 정해진 토큰 세트의 기존 입력 임베딩 값의 학습에 사용되는 복수의 제2 파라미터 그룹, 상기 미리 정해진 토큰 세트의 기존 출력 임베딩 값의 학습에 사용되는 복수의 제3 파라미터 그룹을 고정시킨 상태에서, 미리 정해진 단계 별로 상기 추가 토큰 세트의 추가 입력 임베딩 값의 학습에 사용되는 복수의 제4 파라미터 그룹 또는 상기 추가 토큰 세트의 추가 출력 임베딩 값의 학습에 사용되는 복수의 제5 파라미터 그룹 중 적어도 하나를 학습시키고, 상기 복수의 제1 파라미터 그룹 및 상기 복수의 제2 파라미터 그룹을 고정시킨 상태에서, 미리 정해진 단계 별로 상기 복수의 제3 파라미터 그룹, 상기 복수의 제4 파라미터 그룹, 또는 상기 복수의 제5 파라미터 그룹 중 적어도 하나를 학습시키고, 및 상기 복수의 제1 파라미터 그룹을 고정시키지 않은 상태에서 미리 정해진 단계 별로 상기 복수의 제1 파라미터 그룹 내지 상기 복수의 제5 파라미터 그룹 중 적어도 하나를 학습시키도록 설정될 수 있다. 그 밖의 실시예들도 가능하다. According to various embodiments, a server for performing step-by-step learning for model parameters to realize a multilingual LLM includes a memory, a communication module, and a processor, wherein the processor generates a learning token set by merging an additional token set with a predetermined token set using a pre-learned tokenizer, performs subword-based embedding initialization on the additional token set, and trains at least one of a plurality of fourth parameter groups used for learning additional input embedding values of the additional token set or a plurality of fifth parameter groups used for learning additional output embedding values of the additional token set in a predetermined step-by-step manner while fixing a plurality of first parameter groups used for learning a transformer layer of the multilingual LLM, a plurality of second parameter groups used for learning existing input embedding values of the predetermined token set, and a plurality of third parameter groups used for learning existing output embedding values of the predetermined token set, and while fixing the plurality of first parameter groups and the plurality of second parameter groups, trains at least one of the plurality of third parameter groups, the plurality of fourth parameter groups, or the plurality of fifth parameter groups in a predetermined step-by-step manner. It can be set to train at least one of the parameter groups, and train at least one of the plurality of first parameter groups to the plurality of fifth parameter groups in predetermined steps without fixing the plurality of first parameter groups. Other embodiments are also possible.","assignee":"주식회사 야놀자넥스트","inventors":["김승덕","최승택","정명호"],"publication_date":"2025-10-28","filing_date":"2025-10-13","priority_date":"2024-02-20","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06F","G06F11/00","G06F11/30","G06F11/34","G","G06","G06F","G06F40/00","G06F40/20","G06F40/237","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06F","G06F40/00","G06F40/40","G06F40/53","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250154323A/en"},{"publication_number":"KR20250154315A","title":"Catch-Eye: A Brooch-Type and Expandable Accessory Wearable Device for AI-Based Multilingual Communication and Sign Language Recognition","abstract":"본 발명은 AI 기반 음성 및 수어 인식, 다언어 번역, 홀로그램 디스플레이, 친환경 전력 공급 및 스마트폰 연동 기능을 통합한 악세서리형 웨어러블 디바이스에 관한 것이다. 브로치형 본체에 입출력 통합형 카메라 모듈, AI 프로세서, 홀로그램 디스플레이 모듈 등을 내장하여, 실시간 커뮤니케이션 및 정보 접근성을 지원하며, 태양광 셀 및 무선 충전 시스템을 통해 지속적인 전력 공급이 가능하다. 본 디바이스는 브로치형 외에도 목걸이형, 반지형, 팔찌형, 시계형, 헬멧형 등 다양한 형태로 구조 변경 및 확장이 가능하여, 청각·시각 장애인, 고령자, 외국인, 어린이 등 다양한 사용자 환경에 적응할 수 있는 것을 특징으로 한다. The present invention relates to an accessory-type wearable device that integrates AI-based voice and sign language recognition, multilingual translation, holographic display, eco-friendly power supply, and smartphone linkage functions. The brooch-shaped body incorporates an input/output integrated camera module, AI processor, and hologram display module, enabling real-time communication and information accessibility, and provides continuous power supply through solar cells and a wireless charging system. This device can be structurally changed and expanded into various forms, such as a brooch type, necklace type, ring type, bracelet type, watch type, and helmet type, and is characterized by being adaptable to various user environments, such as the hearing and visual impaired, the elderly, foreigners, and children.","assignee":"유명란","inventors":["유명란"],"publication_date":"2025-10-28","filing_date":"2025-10-09","priority_date":"2025-05-12","cpc_codes":["G","G06","G06F","G06F1/00","G06F1/16","G06F1/1613","G06F1/163","A","A44","A44C","A44C1/00","G","G06","G06F","G06F1/00","G06F1/26","G06F1/32","G","G06","G06F","G06F40/00","G06F40/40","G06F40/58","G","G06","G06N","G06N20/00","H","H02","H02J","H02J3/00","H02J3/38"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250154315A/en"},{"publication_number":"CN120850708A","title":"Automatic prediction method and early warning system for gas storage capacity changes","abstract":"本发明公开了一种储气库库容变化自动预测方法及预警系统，其中方法包括：数据输入与初步处理：从各油井监控系统中采集原始数据，通过初步处理过滤干扰信息并转化为预设格式；数据流动与分支：对初步处理后的数据进行更深层次的数据清洗，清洗后的数据被用于训练横向预测模型与纵向预测模型；模型输出：输入目标井相关数据，选择预测模型或纵向预测模型对目标井库容变化进行预测，并通过可视化图表形式进行呈现，从而判断是否存在库容泄露。本发明通过机器学习预测与实际观测的压力‑注产曲线的对比，可以对储气库库容变化进行精准预测。 The present invention discloses a method and early warning system for automatically predicting changes in gas storage capacity. The method comprises the following steps: data input and preliminary processing: collecting raw data from each oil well monitoring system, filtering out interference information through preliminary processing, and converting it into a preset format; data flow and branching: performing deeper data cleaning on the pre-processed data, and using the cleaned data to train horizontal and vertical prediction models; and model output: inputting data related to a target well, selecting a prediction model or a vertical prediction model to predict changes in the target well's capacity, and presenting the results in a visual chart to determine whether there is a capacity leak. The present invention accurately predicts changes in gas storage capacity by comparing machine learning predictions with actual observed pressure-injection-production curves.","assignee":"Petrochina Co Ltd","inventors":["李力民","王岩","温廷钧","任科","曾娟","钟玉鸣"],"publication_date":"2025-10-28","filing_date":"2024-04-28","priority_date":"2024-04-28","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06F","G06F30/00","G06F30/20","G06F30/28","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06Q","G06Q50/00","G06Q50/02","G","G06","G06F","G06F2111/00","G06F2111/10"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120850708A/en"},{"publication_number":"US12456047B2","title":"Distilling from ensembles to improve reproducibility of neural networks","abstract":"Systems and methods can improve the reproducibility of neural networks by distilling from ensembles. In particular, aspects of the present disclosure are directed to a training scheme that utilizes a combination of an ensemble of neural networks and a single, “wide” neural network that is more powerful (e.g., exhibits a greater accuracy) than the ensemble. Specifically, the output of the ensemble can be distilled into the single neural network during training of the single neural network. After training, the single neural network can be deployed to generate inferences. In such fashion, the single neural model can provide a superior prediction accuracy while, during training, the ensemble can serve to influence the single neural network to be more reproducible. In addition, an additional single wide tower can be added to generate another output, that can be distilled to the single neural network, to further improve its accuracy.","assignee":"Google LLC","inventors":["Gil Shamir","Lorenzo Coviello"],"publication_date":"2025-10-28","filing_date":"2020-09-18","priority_date":"2019-11-21","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12456047B2/en"},{"publication_number":"US12456034B2","title":"Image classification explanation by generating boundary crossing examples with removed features via filter suppression","abstract":"A computer-implemented method for explaining a classification of one or more classifier inputs by a trained classifier. A generative model is used that generates inputs for the trained classifier. The generative model comprises multiple filters. Generator inputs corresponding to the one or more classifier inputs are obtained, where a generator input causes the generative model to approximately generate the corresponding classifier input. Filter suppression factors are determined for the multiple filters of the generative model. A filter suppression factor for a filter indicates a degree of suppression for a filter output of the filter. The filter suppression factors are determined based on an effect of adapting the classifier inputs according to the filter suppression factors on the classification by the trained classifier. The classification explanation is based on the filter suppression factors.","assignee":"Robert Bosch GmbH","inventors":["Andres Mauricio Munoz Delgado"],"publication_date":"2025-10-28","filing_date":"2021-04-13","priority_date":"2020-04-20","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N5/00","G06N5/04","G06N5/045"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12456034B2/en"},{"publication_number":"CN120852089A","title":"A data knowledge base management method, device and medium for the egg industry","abstract":"The invention discloses a data knowledge base management method, equipment and medium for egg industry, which relate to the technical field of intelligent culture informatization and comprise the steps of inputting a multi-mode fusion data set into a single deep learning frame, carrying out multi-scale feature coupling on each mode feature through a self-adaptive nonlinear interaction function, dynamically adjusting feature weights by combining an enhanced feedback mechanism, generating a multi-dimensional health assessment index set by utilizing end-to-end combined feature extraction and nonlinear reasoning, carrying out entity identification and relation extraction on the multi-dimensional health assessment index set, combining culture environment and individual feature information, constructing an egg industry knowledge map, carrying out structural optimization and dynamic evolution to obtain a dynamic evolution knowledge map, and carrying out rule reasoning and data analysis on the dynamic evolution knowledge map to generate a health management decision. The precision modeling and high-quality unified expression of the multi-source data are realized through piecewise nonlinear feature transformation and weighted fusion, and the accuracy of the health state analysis input is ensured.","assignee":"Chengcheng Zhishu Technology Shenzhen Co ltd","inventors":["黎超勇","刘冬青","詹锦涛","詹少妹"],"publication_date":"2025-10-28","filing_date":"2025-09-03","priority_date":"2025-09-03","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/02","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06N","G06N5/00","G06N5/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120852089A/en"},{"publication_number":"CN120850182A","title":"Anomaly root cause identification method for multivariate time series data based on spatiotemporal causal graph","abstract":"本发明涉及一种基于时空因果图的多元时序数据异常根因识别方法，属于异常检测技术领域。所述方法，对多元时序数据，采用多窗口膨胀因果卷积，结合互信息筛选，提取同时覆盖短期突变与长时依赖的时间嵌入；利用多头自注意力学习非局部空间关联，再以条件熵度量方向性因果强度，并通过归一化‑剪枝生成稀疏、可解释的时空因果图；在时空因果图上引入因果增强图注意力网络，对节点嵌入进行多轮因果传播更新；综合异常程度与因果影响力计算根因评分，识别出异常传播路径中的关键源节点，实现系统异常的精准根因定位。本发明增强了对动态行为模式的适应性和对异常驱动因子的捕捉能力，提升了异常传播过程的建模精度与根因识别能力。 The present invention relates to a method for identifying the root cause of anomalies in multivariate time series data based on a spatiotemporal causal graph, and belongs to the field of anomaly detection technology. The method adopts multi-window dilated causal convolution for multivariate time series data, combined with mutual information screening, to extract time embeddings that simultaneously cover short-term mutations and long-term dependencies; utilizes multi-head self-attention to learn non-local spatial associations, and then measures the directional causal strength with conditional entropy, and generates a sparse and interpretable spatiotemporal causal graph through normalization-pruning; introduces a causal enhancement graph attention network on the spatiotemporal causal graph, and performs multiple rounds of causal propagation updates on node embeddings; calculates the root cause score by combining the degree of anomaly and causal influence, identifies the key source nodes in the anomaly propagation path, and realizes the precise root cause location of system anomalies. The present invention enhances the adaptability to dynamic behavior patterns and the ability to capture anomaly driving factors, and improves the modeling accuracy and root cause identification capability of the anomaly propagation process.","assignee":"Fujian Normal University","inventors":["许力","曾钻洋","李家印","叶阿勇","林丽美","汪晓丁","张柳明"],"publication_date":"2025-10-28","filing_date":"2025-09-24","priority_date":"2025-09-24","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2433","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06F","G06F2123/00","G06F2123/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120850182A/en"},{"publication_number":"CN115422945B","title":"A rumor detection method and system integrating sentiment mining","abstract":"本发明提出一种融合情感挖掘的谣言检测方法，所述方法包括以下步骤；步骤A：收集并提取社交网络媒体中源帖子的文本内容和评论内容，并人工标注源帖子的真实标签，形成训练数据集DT；步骤B：使用训练数据集DT，训练基于多级注意力和知识图谱的深度学习网络模型N，训练内容包括分析源帖子的真实性和预测源帖子的真实性标签；步骤C：将源帖子的文本内容和评论内容输入到训练好的深度学习网络模型N中，获得源帖子的真实性标签；本发明可以提升对微博进行谣言检测的准确性。 The present invention proposes a rumor detection method that integrates sentiment mining, and the method comprises the following steps: Step A: collecting and extracting the text content and comment content of source posts in social network media, and manually marking the real labels of the source posts to form a training data set DT; Step B: using the training data set DT, training a deep learning network model N based on multi-level attention and knowledge graph, and the training content includes analyzing the authenticity of the source posts and predicting the authenticity labels of the source posts; Step C: inputting the text content and comment content of the source posts into the trained deep learning network model N to obtain the authenticity labels of the source posts; the present invention can improve the accuracy of rumor detection on Weibo.","assignee":"Fuzhou University","inventors":["陈羽中","朱文龙","饶孟宇","万宇杰"],"publication_date":"2025-10-28","filing_date":"2022-09-19","priority_date":"2022-09-19","cpc_codes":["G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06F","G06F16/00","G06F16/30","G06F16/36","G06F16/367","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G06F40/211","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN115422945B/en"},{"publication_number":"CN120851193A","title":"A multi-agent social network simulation method and system based on cognitive reasoning chain","abstract":"本发明涉及人工智能技术领域，公开了一种基于认知推理链的多智能体社交网络模拟方法及系统，方法包括：初始化包括社会环境引擎、用户画像引擎和认知推理引擎的多智能体系统，执行预设轮次迭代的多智能体社交网络模拟循环，在每个迭代轮次中社会环境引擎向智能体推送社交信息作为外部刺激并激活智能体执行独立决策，通过大型语言模型推理更新认知推理链中各维度的认知状态并生成相应的社交行为，记录社交行为及对应的认知状态轨迹；周期性分析历史记录来优化认知推理链各认知维度间的影响系数，调整预设大型语言模型的推理策略。使得对智能体认知过程的模拟为一个透明、可追踪的演化过程，实现了对“观察–认知–行为”周期的完整且可理解的模拟。 The present invention relates to the field of artificial intelligence technology and discloses a multi-agent social network simulation method and system based on a cognitive reasoning chain. The method comprises: initializing a multi-agent system comprising a social environment engine, a user profiling engine, and a cognitive reasoning engine; executing a multi-agent social network simulation loop with preset iterations; in each iteration, the social environment engine pushes social information to the agents as external stimuli and activates the agents to make independent decisions; updating the cognitive states of each dimension in the cognitive reasoning chain through reasoning with a large language model and generating corresponding social behaviors; and recording the social behaviors and corresponding cognitive state trajectories; and periodically analyzing historical records to optimize the influence coefficients between the cognitive dimensions in the cognitive reasoning chain and adjust the reasoning strategy of the preset large language model. This method makes the simulation of the agent's cognitive process a transparent and traceable evolutionary process, achieving a complete and understandable simulation of the \"observation-cognition-behavior\" cycle.","assignee":"Harbin Institute Of Technology shenzhen Shenzhen Institute Of Science And Technology Innovation Harbin Institute Of Technology","inventors":["廖清","钟林","汪灵芝"],"publication_date":"2025-10-28","filing_date":"2025-06-27","priority_date":"2025-06-27","cpc_codes":["H","H04","H04L","H04L51/00","H04L51/52","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06N","G06N7/00","G06N7/01"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120851193A/en"},{"publication_number":"KR20250153744A","title":"Apparatus and method for lightning prediction using artificial neural network","abstract":"낙뢰의 발생 확률, 이동 속도 및 방향을 예측할 수 있는 낙뢰 예측 장치 및 방법이 개시된다. 일 실시예에 따른 낙뢰 예측 장치는 입력 데이터를 수신하여 공간적 및 시간적 데이터 전처리를 수행하는 전처리부; 상기 전처리된 입력 데이터를 기초로 소정 시간 이후 2차원 격자 별 낙뢰 발생 확률을 계산하여 낙뢰 예측 지도를 생성하는 낙뢰 예측부; 및 상기 낙뢰 예측 지도에 기초하여 낙뢰 중심점을 계산하며, 소정 시간 간격에 대한 복수의 낙뢰 중심점을 기초로 이동 방향 및 속도를 계산하는 이동 예측부를 포함할 수 있다. A lightning prediction device and method capable of predicting the occurrence probability, movement speed, and direction of lightning are disclosed. According to one embodiment, the lightning prediction device may include a preprocessing unit that receives input data and performs spatial and temporal data preprocessing; a lightning prediction unit that calculates the occurrence probability of lightning for each two-dimensional grid after a predetermined time based on the preprocessed input data to generate a lightning prediction map; and a movement prediction unit that calculates a lightning center point based on the lightning prediction map and calculates a movement direction and speed based on a plurality of lightning center points for a predetermined time interval.","assignee":"(주)한국해양기상기술","inventors":["이원용","하지훈"],"publication_date":"2025-10-27","filing_date":"2025-10-14","priority_date":"2023-12-28","cpc_codes":["G","G01","G01W","G01W1/00","G01W1/10","G","G01","G01S","G01S13/00","G01S13/88","G01S13/95","G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G01","G01W","G01W2201/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250153744A/en"},{"publication_number":"KR20250153738A","title":"Methods and apparatus for compressing data streams","abstract":"데이터 스트림들을 압축하기 위한 방법들 및 장치가 개시된다. 일 실시예에서, 방법은, 스칼라 데이터에 대한 확률 분포 함수(PDF)를 계산하는 단계와, 가장 근접한 매칭 PDF 템플릿을 결정하기 위해 상기 PDF를 PDF 템플릿들과 매칭하는 단계와, 그리고 대응하는 인코더 식별자가 결정되는 상기 가장 근접한 매칭 PDF 템플릿에 대응하는 인코더를 선택하는 단계를 포함한다. 상기 방법은 또한 인코딩된 스트림을 생성하기 위해 상기 스칼라 데이터를 상기 인코더로 인코딩하는 단계와; 그리고 상기 인코딩된 스트림 및 상기 인코더 식별자를 전송하는 단계를 포함한다. Methods and apparatus for compressing data streams are disclosed. In one embodiment, the method comprises the steps of computing a probability distribution function (PDF) for scalar data, matching the PDF with PDF templates to determine a closest matching PDF template, and selecting an encoder corresponding to the closest matching PDF template, wherein a corresponding encoder identifier is determined. The method also comprises the steps of encoding the scalar data with the encoder to generate an encoded stream; and transmitting the encoded stream and the encoder identifier.","assignee":"마벨 아시아 피티이 엘티디","inventors":["칼펜두 라탄쉬 파사드","김홍직"],"publication_date":"2025-10-27","filing_date":"2025-10-10","priority_date":"2019-04-30","cpc_codes":["H","H03","H03M","H03M7/00","H03M7/30","H03M7/40","H","H04","H04L","H04L69/00","H04L69/04","H","H03","H03M","H03M7/00","H03M7/30","H03M7/60","H03M7/6011","G","G06","G06N","G06N7/00","G06N7/01","H","H03","H03M","H03M7/00","H03M7/30","H","H03","H03M","H03M7/00","H03M7/30","H03M7/3084","H","H03","H03M","H03M7/00","H03M7/30","H03M7/60","H03M7/6017","H","H04","H04L","H04L47/00","H04L47/10","H04L47/12","H","H03","H03M","H03M7/00","H03M7/30","H03M7/3082"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250153738A/en"},{"publication_number":"KR20250153721A","title":"Method and system for generating answers","abstract":"본 발명은, 답변 생성 방법 및 시스템에 관한 것이다. 보다 구체적으로, 본 발명에 따른, 메모리 및 적어도 하나의 프로세서가 협력하여 수행되는, 답변 생성 방법은, 문서 이해(Document Understanding) 모델을 이용하여, 분석 대상 문서로부터, 적어도 하나의 분자 구조를 추출하는 단계, 상기 메모리에, 상기 문서로부터 추출된 분자 구조를 저장하는 단계, 상기 메모리에 저장된 상기 추출된 분자 구조 마다 서로 다른 라벨이 부여되도록, 상기 추출된 분자 구조에 대한 라벨링을 수행하는 단계, 서비스 페이지를 통해, 상기 라벨링을 통해 부여된 라벨 중 적어도 하나를 포함하는 사용자 질의를 수신하는 단계 및 상기 추출된 분자 구조 중 상기 사용자 질의에 포함된 특정 라벨에 대응되는 분자 구조를 이용하여, 상기 사용자 질의에 대한 답변을 생성하는 단계를 포함할 수 있다. The present invention relates to a method and system for generating an answer. More specifically, the method for generating an answer, which is performed cooperatively by a memory and at least one processor according to the present invention, may include the steps of extracting at least one molecular structure from a document to be analyzed using a document understanding model, storing the molecular structure extracted from the document in the memory, performing labeling on the extracted molecular structure so that a different label is assigned to each of the extracted molecular structures stored in the memory, receiving a user query including at least one of the labels assigned through the labeling through a service page, and generating an answer to the user query using a molecular structure corresponding to a specific label included in the user query among the extracted molecular structures.","assignee":"주식회사 Lg 경영개발원","inventors":["호르마자발 로드리고","베르턴스 폴","한세희"],"publication_date":"2025-10-27","filing_date":"2025-09-30","priority_date":"2023-07-19","cpc_codes":["G","G16","G16C","G16C20/00","G16C20/70","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G16","G16C","G16C20/00","G16C20/10","G","G16","G16C","G16C20/00","G16C20/40","G","G16","G16C","G16C20/00","G16C20/90"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250153721A/en"},{"publication_number":"KR102876286B1","title":"AI based prompt automatic generation system and method using context information in development tool and program performing the same","abstract":"개발도구에서 맥락 정보를 활용한 AI 기반 프롬프트 자동 생성 시스템 및 방법과 이를 수행하는 프로그램이 개시된다. 본 발명의 일 측면에 따르면, AI 기반 프롬프트 자동 생성 방법을 수행하도록 하기 위해 컴퓨터-판독 가능 매체에 저장된 컴퓨터 프로그램으로서, 상기 컴퓨터 프로그램은 컴퓨터로 하여금 이하의 단계들을 수행하도록 하며, 상기 단계들은, 개발도구를 통한 앱 개발에 관련된 현재 개발 환경과 작업 내역에 관한 맥락 정보를 수집하는 단계; 상기 맥락 정보를 바탕으로 맥락을 분석하는 단계; 맥락 분석 결과에 기초하여 LLM에게 질의할 후보 프롬프트를 자동 생성하는 단계; 대화형 UI를 통해 상기 후보 프롬프트를 제공하는 단계; 상기 후보 프롬프트에 대한 사용자 선택이 입력되면, 사용자 선택 결과에 따른 프롬프트를 AI 지원 시스템으로 전달하여 LLM에게 제공하고, 응답을 수신하게 하는 단계를 포함하는 컴퓨터-판독 가능 매체에 저장된 컴퓨터 프로그램이 제공된다. A system and method for automatically generating AI-based prompts using contextual information in a development tool and a program for performing the same are disclosed. According to one aspect of the present invention, a computer program stored in a computer-readable medium is provided for performing the AI-based automatic prompt generation method, wherein the computer program causes a computer to perform the following steps, the steps including: collecting contextual information about a current development environment and work history related to app development using the development tool; analyzing a context based on the contextual information; automatically generating a candidate prompt to be queried to an LLM based on the contextual analysis result; providing the candidate prompt through an interactive UI; and, when a user selection for the candidate prompt is input, transmitting a prompt based on the user selection result to an AI-supported system to provide the prompt to the LLM and receiving a response.","assignee":"주식회사 인스웨이브","inventors":["어세룡","김욱래"],"publication_date":"2025-10-27","filing_date":"2025-07-01","priority_date":"2025-07-01","cpc_codes":["G","G06","G06F","G06F8/00","G06F8/30","G06F8/33","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F40/00","G06F40/30","G06F40/35","G","G06","G06F","G06F8/00","G06F8/30","G06F8/38","G","G06","G06N","G06N20/00"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102876286B1/en"},{"publication_number":"KR102876021B1","title":"Method for managing community facilities using artificial intelligence and electronic device for performing the same","abstract":"전자 장치는 자연어 질의를 수신하고, 상기 자연어 질의를 분석하여 질의 의도 예측, 질의 슬롯 추출, 필수 슬롯 누락 여부 검출, 및 부족 슬롯에 대한 추가 질문을 생성하고, 상기 추가 질문에 대한 응답을 수신하면 상기 자연어 질의 및 상기 응답에 기초하여 키워드를 생성하는 자연어 처리부; 상기 자연어 처리부로부터 출력된 상기 키워드를 변환하여 페이로드를 생성하는 프로토콜 변환부; 상기 프로토콜 변환부로부터 전달된 상기 페이로드에 기초하여 원시 데이터를 획득하는 데이터 수집부; 및 상기 데이터 수집부로부터 수신된 원시 데이터를 처리하여 최종 데이터를 생성하는 데이터 생성부를 포함한다. An electronic device includes a natural language processing unit that receives a natural language query, analyzes the natural language query to predict the intent of the query, extract a query slot, detect whether a required slot is missing, and generate an additional question for a missing slot, and generates a keyword based on the natural language query and the response when a response to the additional question is received; a protocol conversion unit that converts the keyword output from the natural language processing unit to generate a payload; a data collection unit that obtains raw data based on the payload transmitted from the protocol conversion unit; and a data generation unit that processes the raw data received from the data collection unit to generate final data.","assignee":"주식회사 트러스테이","inventors":["이승오","임태민","배영기","임현철","김영환","이정현","박영호"],"publication_date":"2025-10-27","filing_date":"2025-05-13","priority_date":"2025-05-13","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3344","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/338","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06Q","G06Q50/00","G06Q50/10"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102876021B1/en"},{"publication_number":"KR102876384B1","title":"The Method, System, and Computer-readable Storage Medium Of OCR Using Encoder And Decoder Based On Artificial Neural Network","abstract":"본 발명은 인공신경망 기반의 인코더 및 디코더를 이용한 OCR 방법, 시스템, 및 컴퓨터-판독가능 저장매체로써, 더 구체적으로는, 이미지를 인코더모델에 입력하여 특징정보를 도출하고 특징정보를 디코더모델에 입력하여 음절 단위의 토큰시퀀스를 도출한 뒤, 토큰시퀀스에 음절-토큰매핑맵을 적용하여 토큰시퀀스에 해당하는 텍스트를 도출하는, 인공신경망 기반의 인코더 및 디코더를 이용한 OCR 방법, 시스템, 및 컴퓨터-판독가능 저장매체에 관한 것이다. The present invention relates to an OCR method, system, and computer-readable storage medium using an encoder and decoder based on an artificial neural network, and more specifically, to an OCR method, system, and computer-readable storage medium using an encoder and decoder based on an artificial neural network, which inputs an image into an encoder model to derive feature information, inputs the feature information into a decoder model to derive a token sequence in syllable units, and then applies a syllable-token mapping map to the token sequence to derive text corresponding to the token sequence.","assignee":"셀렉트스타 주식회사","inventors":["심민우","김남길","강바롬","정준녕"],"publication_date":"2025-10-27","filing_date":"2025-04-28","priority_date":"2025-04-28","cpc_codes":["G","G06","G06V","G06V30/00","G06V30/10","G06V30/18","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V30/00","G06V30/10","G06V30/19","G06V30/191","G06V30/19127"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102876384B1/en"},{"publication_number":"LU601354B1","title":"Automated anomaly monitoring system based on artificial intelligence","abstract":"An automated anomaly monitoring system based on artificial intelligence, comprising an space-air-ground integrated information collection module, a multi-modal data processing module, an automated anomaly monitoring model design module, a model performance enhancement module and a data automated anomaly classification module. The invention relates to the technical field of abnormal data processing; this solution innovatively proposes a three-layer collaborative perception information collection architecture of air-space-ground, which has the advantages of full-scene coverage, multi-granularity perception, and strong spatiotemporal continuit in data collection, enhancing the data integrity and reliability of abnormal event monitoring; it innovatively designs a cross-modal spatiotemporal fusion analysis model, realizing the deep spatiotemporal fusion of multi-source heterogeneous data, improving the accuracy of abnormal event monitoring; it innovatively improves the particle swarm algorithm through a dynamic mutation rate control strategy to obtain the optimal model parameters, thereby improving the response speed and monitoring accuracy of abnormal monitoring.","assignee":"Zhaojia Yang","inventors":["Zhaojia Yang"],"publication_date":"2025-10-27","filing_date":"2025-04-25","priority_date":"2025-04-25","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/10","G06V20/13","G","G06","G06N","G06N3/00","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/10","G06V20/17","G","G08","G08G","G08G5/00","H","H04","H04W","H04W4/00","H04W4/30","H04W4/38"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU601354B1/en"},{"publication_number":"KR102877746B1","title":"Apparatus and method for providing microgrid resource management service based on artificial intelligence","abstract":"본 발명은 인공지능 기반의 마이크로그리드 자원 관리 서비스 제공 장치 및 방법에 관한 것이다. 본 발명의 일 실시예에 따른 전자 장치는 메모리(memory) 및 상기 메모리와 연결된 프로세서(processor)를 포함하고, 상기 프로세서는, 재생에너지생성및저장장치가 위치하는 대상지역의 기상데이터를 수집하고, 상기 기상데이터를 기반으로 에너지의 발전량을 예측하는 기상분석모듈을 통하여 제1 분석데이터를 생성하고, 상기 대상지역의 전력생산 및 전력소비데이터를 수집하고, 상기 대상지역의 에너지부하 및 수요패턴을 예측하는 부하및수요분석모듈을 통하여 제2 분석데이터를 생성하고, 상기 재생에너지생성및저장장치에 포함되는 에너지저장장치의 에너지 단위당 비용을 도출하는 저장비용계산모듈을 통하여 제3 분석데이터를 생성하고, 제1 인공지능모듈을 통하여, 상기 제1 분석데이터, 상기 제2 분석데이터 및 상기 제3 분석데이터를 기반으로 상기 재생에너지생성및저장장치의 제어명령을 생성하고, 상기 제어명령을 상기 재생에너지생성및저장장치에게 송신할 수 있다. The present invention relates to an apparatus and method for providing an artificial intelligence-based microgrid resource management service. An electronic device according to one embodiment of the present invention includes a memory and a processor connected to the memory, wherein the processor collects meteorological data of a target area where a renewable energy generation and storage device is located, generates first analysis data through a meteorological analysis module that predicts the amount of energy generated based on the meteorological data, generates second analysis data through a load and demand analysis module that collects power generation and power consumption data of the target area and predicts energy load and demand patterns of the target area, generates third analysis data through a storage cost calculation module that derives a cost per energy unit of an energy storage device included in the renewable energy generation and storage device, and generates a control command of the renewable energy generation and storage device based on the first analysis data, the second analysis data, and the third analysis data through a first artificial intelligence module, and transmits the control command to the renewable energy generation and storage device.","assignee":"렉스이노베이션 주식회사","inventors":["코일존","임정민"],"publication_date":"2025-10-27","filing_date":"2025-03-09","priority_date":"2025-03-09","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/06","G","G01","G01W","G01W1/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G08","G08B","G08B21/00","G08B21/18","G08B21/185","H","H02","H02J","H02J3/00","H02J3/28","H02J3/32","Y","Y04","Y04S","Y04S10/00","Y04S10/12","Y04S10/123"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102877746B1/en"},{"publication_number":"KR102876030B1","title":"Method for removing artifact of medical image using artificial intelligence and system for the same","abstract":"혈관의 OCT(Optical Coherence Tomography) 영상에서 아티펙트를 제거하는 방법에 있어서, 혈관에 대한 OCT 이미지를 획득하는 단계; 상기 OCT 이미지에 포함된 아티펙트 영역을 탐지하는 단계; 상기 OCT 이미지 및 상기 아티펙트 영역에 대한 정보를 제1인공지능 모델에 입력하여, 추론 이미지를 생성하는 단계; 및 상기 생성된 추론 이미지를 후처리하여 최종 OCT 이미지를 생성하는 단계; 를 포함할 수 있다. A method for removing artifacts from an OCT (Optical Coherence Tomography) image of a blood vessel may include: a step of acquiring an OCT image of the blood vessel; a step of detecting an artifact region included in the OCT image; a step of inputting information about the OCT image and the artifact region into a first artificial intelligence model to generate an inference image; and a step of post-processing the generated inference image to generate a final OCT image.","assignee":"노바사이언스 주식회사","inventors":["김원진","조정애","이찬미"],"publication_date":"2025-10-27","filing_date":"2024-12-24","priority_date":"2024-12-24","cpc_codes":["G","G16","G16H","G16H30/00","G16H30/40","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06T","G06T5/00","G06T5/50","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G16","G16H","G16H30/00","G16H30/20","G","G16","G16H","G16H50/00","G16H50/50","G","G16","G16H","G16H50/00","G16H50/70","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10072","G06T2207/10101","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30004","G06T2207/30021"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102876030B1/en"},{"publication_number":"KR102877638B1","title":"Method and apparatus for neural network quantization","abstract":"뉴럴 네트워크 양자화를 위한 방법 및 장치는, 뉴럴 네트워크의 학습을 반복적으로 수행하고, 뉴럴 네트워크에 포함된 레이어들 각각의 웨이트 통계량을 분석하고, 분석된 통계량에 기초하여 낮은 비트 정밀도로 양자화될 레이어들을 결정하고, 결정된 레이어들을 낮은 비트 정밀도로 양자화함으로써 양자화된 뉴럴 네트워크를 생성한다. A method and device for quantizing a neural network repeatedly perform training of a neural network, analyze weight statistics of each layer included in the neural network, determine layers to be quantized with low bit precision based on the analyzed statistics, and quantize the determined layers with low bit precision, thereby generating a quantized neural network.","assignee":"삼성전자주식회사","inventors":["이원조","이승원","이준행"],"publication_date":"2025-10-27","filing_date":"2019-01-09","priority_date":"2019-01-09","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102877638B1/en"},{"publication_number":"KR20250153882A","title":"Methods and mechanisms for adjusting film deposition parameters during substrate manufacturing","abstract":"전자 디바이스 제조 시스템은 프로세스 레시피에 따라 기판 상에 수행되는 퇴적 프로세스에 연관된 계측 데이터를 획득할 수 있고, 퇴적 프로세스는 기판의 표면 상에 복수의 층을 생성한다. 제조 시스템은 추가로, 프로세스 레시피에 연관된 예상 프로파일을 획득할 수 있고, 예상 프로파일은 프로세스 레시피의 복수의 층에 대한 원하는 두께를 나타내는 복수의 값을 포함한다. 제조 시스템은 추가로, 계측 데이터 및 예상 프로파일에 기초하여 수정 프로파일을 생성할 수 있고, 수정 프로파일은 복수의 층 중 적어도 하나의 층에 대한 퇴적 시간 오프셋 값을 포함한다. 제조 시스템은 추가로, 수정 프로파일을 프로세스 레시피에 적용함으로써 업데이트된 프로세스 레시피를 생성할 수 있고, 업데이트된 프로세스 레시피에 따라 기판 상에서 퇴적 단계가 수행되게 할 수 있다. An electronic device manufacturing system can obtain measurement data associated with a deposition process performed on a substrate according to a process recipe, wherein the deposition process creates a plurality of layers on a surface of the substrate. The manufacturing system can further obtain an expected profile associated with the process recipe, wherein the expected profile includes a plurality of values representing desired thicknesses for the plurality of layers of the process recipe. The manufacturing system can further generate a modified profile based on the measurement data and the expected profile, wherein the modified profile includes a deposition time offset value for at least one of the plurality of layers. The manufacturing system can further generate an updated process recipe by applying the modified profile to the process recipe, and cause a deposition step to be performed on the substrate according to the updated process recipe.","assignee":"어플라이드 머티어리얼스, 인코포레이티드","inventors":["미테쉬 상비","벤카타나라야나 샨카라무르티","유렌 야오","추안 잉 왕","신하이 한"],"publication_date":"2025-10-27","filing_date":"2023-05-04","priority_date":"2022-05-05","cpc_codes":["G","G05","G05B","G05B19/00","G05B19/02","G05B19/418","C","C23","C23C","C23C16/00","C23C16/44","C23C16/455","C23C16/45523","C23C16/45525","C","C23","C23C","C23C16/00","C23C16/44","C23C16/52","G","G05","G05B","G05B19/00","G05B19/02","G05B19/418","G05B19/41865","G","G06","G06N","G06N20/00","G","G05","G05B","G05B2219/00","G05B2219/30","G05B2219/31","G05B2219/31443","G","G05","G05B","G05B2219/00","G05B2219/30","G05B2219/45","G05B2219/45031"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250153882A/en"},{"publication_number":"KR20250153165A","title":"Method and apparatus for generating 3-dimensional motion video based on artificial intelligence","abstract":"인공지능 기반의 3차원 모션 영상 생성 방법 및 장치가 제공된다. 본 발명의 일 실시예에 따른, 인공지능 기반의 3차원 모션 영상 생성 방법은 컴퓨팅 장치에 의해 수행되고, 2차원 영상을 획득하는 단계, 기계학습 된 인공지능 모듈을 이용하여 상기 2차원 영상으로부터 객체를 식별하고 상기 식별된 객체의 움직임을 분석하여 상기 객체의 3차원 모델링 데이터를 생성하는 단계, 및 상기 3차원 모델링 데이터를 이용하여 3차원 모션 영상을 생성하는 단계를 포함할 수 있다. 본 발명에 따른 3차원 모션 영상 생성 방법에 의하면, 인공지능 기반으로 2차원 영상으로부터 객체의 동작을 추출하여 이를 3차원 모션 영상으로 구현할 수 있다. 또한, 객체의 2차원 영상을 이용하여 상기 객체의 3차원 메쉬 데이터를 생성할 수 있다. A method and device for generating a three-dimensional motion image based on artificial intelligence are provided. According to one embodiment of the present invention, a method for generating a three-dimensional motion image based on artificial intelligence is performed by a computing device and may include a step of acquiring a two-dimensional image, a step of identifying an object from the two-dimensional image using a machine-learned artificial intelligence module and analyzing the movement of the identified object to generate three-dimensional modeling data of the object, and a step of generating a three-dimensional motion image using the three-dimensional modeling data. According to the method for generating a 3D motion image according to the present invention, the motion of an object can be extracted from a 2D image based on artificial intelligence and implemented as a 3D motion image. In addition, 3D mesh data of the object can be generated using the 2D image of the object.","assignee":"주식회사 네이션에이","inventors":["유수연","최정원"],"publication_date":"2025-10-24","filing_date":"2025-10-14","priority_date":"2022-07-22","cpc_codes":["G","G06","G06T","G06T13/00","G06T13/20","G","G06","G06T","G06T7/00","G06T7/50","G06T7/55","G06T7/593","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06T","G06T11/00","G06T11/60","G","G06","G06T","G06T13/00","G06T13/20","G06T13/40","G","G06","G06T","G06T13/00","G06T13/80","G","G06","G06T","G06T17/00","G","G06","G06T","G06T17/00","G06T17/20","G","G06","G06T","G06T19/00","G","G06","G06T","G06T19/00","G06T19/20","G","G06","G06T","G06T5/00","G","G06","G06T","G06T5/00","G06T5/60","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G06","G06T","G06T7/00","G06T7/10","G06T7/194","G","G06","G06T","G06T7/00","G06T7/20"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250153165A/en"},{"publication_number":"KR20250153160A","title":"System for observing intruding objects for protecting electric power facilities and method for respondng","abstract":"전력 설비 보호를 위한 침입 객체 감시 시스템 및 이를 이용한 대응 방법을 제공한다. 전력 설비의 침입 객체에 대한 대응 방법은, i) 영상 입력에 따라 인공지능 모델부가 침입 객체를 필터링하여 식별하는 단계, ii) 침입 객체가 사람, 동물 및 조류로 이루어진 군에서 선택된 하나 이상의 생물체인지 여부를 판단하는 단계, iii) 생물체를 전력 설비의 관리자에게 통지하는 단계, 및 iv) 생물체가 전력 설비 중 주요 설비를 침범한 것으로 판단되는 경우, 경보를 발생시키는 단계를 포함한다. A system for monitoring intrusion objects for power facility protection and a method for responding using the same are provided. The method for responding to an intrusion object in a power facility includes: i) a step of having an artificial intelligence model section filter and identify an intrusion object based on an image input; ii) a step of determining whether the intrusion object is one or more living organisms selected from a group consisting of humans, animals, and birds; iii) a step of notifying a manager of the power facility of the living organism; and iv) a step of generating an alarm if it is determined that the living organism has intruded into a major facility among the power facilities.","assignee":"한국전력공사","inventors":["김광재","윤여형"],"publication_date":"2025-10-24","filing_date":"2025-10-10","priority_date":"2022-08-30","cpc_codes":["G","G08","G08B","G08B13/00","G08B13/18","G08B13/189","G08B13/194","G08B13/196","G08B13/19602","A","A01","A01M","A01M29/00","A01M29/16","A01M29/18","F","F21","F21V","F21V23/00","F21V23/04","F21V23/0442","F21V23/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06V","G06V10/00","G06V10/10","G06V10/12","G06V10/14","G06V10/141","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V40/00","G06V40/10","G","G08","G08B","G08B25/00","G08B25/14","G","G08","G08B","G08B3/00","G08B3/10","H","H04","H04N","H04N7/00","H04N7/18","H04N7/181","F","F21","F21W","F21W2131/00","F21W2131/40"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250153160A/en"},{"publication_number":"CN120831384A","title":"Gas leakage quantitative calibration method and system based on multi-spectral feature fusion","abstract":"本申请涉及红外热成像的技术领域，尤其是涉及一种基于多光谱特征融合的气体泄漏定量标定方法及系统，其包括：配置多波段黑体辐射源；搭建模拟气体云团环境并实时获取配置参数，并对配置参数进行分类采集以构建四维参数空间；获取针对模拟气体云团环境所对应的红外热成像数据，并与四维参数空间进行关联以形成结构化样本集；利用双通道3D‑CNN共享结构化样本集以提取辐射强度特征和云团形态特征并进行拼接以得到融合特征向量；对融合特征向量进行基于深度学习的非线性映射以输出浓度预测值；引入大气传输校正因子对浓度预测值进行修正以得到浓度精确值。本申请具有提高气体泄漏浓度标定的准确性的效果。 The present application relates to the technical field of infrared thermal imaging, and in particular to a method and system for quantitatively calibrating gas leaks based on multispectral feature fusion. The method comprises: configuring a multi-band blackbody radiation source; constructing a simulated gas cloud environment and acquiring configuration parameters in real time, and classifying and collecting the configuration parameters to construct a four-dimensional parameter space; acquiring infrared thermal imaging data corresponding to the simulated gas cloud environment and associating it with the four-dimensional parameter space to form a structured sample set; utilizing a dual-channel 3D-CNN to share the structured sample set to extract radiation intensity characteristics and cloud morphological characteristics and to concatenate them to obtain a fused feature vector; performing a deep learning-based nonlinear mapping on the fused feature vector to output a concentration prediction value; and introducing an atmospheric transmission correction factor to correct the concentration prediction value to obtain an accurate concentration value. The present application has the effect of improving the accuracy of gas leak concentration calibration.","assignee":"Zhejiang Hongpu Inc","inventors":["邓丰涛","钟小露","王剑","李东叶","周涛","李茂宇"],"publication_date":"2025-10-24","filing_date":"2025-09-22","priority_date":"2025-09-22","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G01","G01J","G01J5/00","G01J5/48","G","G01","G01J","G01J5/00","G01J5/90","G","G01","G01M","G01M3/00","G01M3/002","G","G01","G01M","G01M3/00","G01M3/007","G","G01","G01N","G01N25/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120831384A/en"},{"publication_number":"CN120833557A","title":"Construction site image processing method, device, equipment and medium","abstract":"本发明涉及图像处理技术领域，提供一种施工现场图像处理方法、装置、设备及介质，能够识别目标施工现场图像中的非目标物体，并根据识别结果在目标施工现场图像中生成掩膜，先进行粗略的非目标物体定位，再根据定位进行精确地目标分割，能够有效平衡处理效率与精度；根据掩膜对目标施工现场图像进行擦除处理得到带有空白区域的第一中间图像，基于环境感知算法对空白区域进行内容补全得到第二中间图像，能够解决动态干扰物导致的特征匹配失效问题，同时确保建筑主体结构完整性；基于建筑信息模型的空间语义信息对第二中间图像进行细节还原处理，能够降低光照变化与遮挡干扰的影响，从而生成更加适用于执行施工现场监测任务的高质量图像。 The present invention relates to the field of image processing technology, and provides a construction site image processing method, device, equipment and medium. The methods are capable of identifying non-target objects in a target construction site image, and generating a mask in the target construction site image based on the identification result. The non-target objects are first roughly located, and then the target is accurately segmented based on the location, which can effectively balance processing efficiency and accuracy. The target construction site image is erased according to the mask to obtain a first intermediate image with a blank area, and the blank area is supplemented with content based on an environmental perception algorithm to obtain a second intermediate image, which can solve the problem of feature matching failure caused by dynamic interference objects while ensuring the integrity of the main building structure. The second intermediate image is restored with details based on the spatial semantic information of a building information model, which can reduce the influence of illumination changes and occlusion interference, thereby generating a high-quality image that is more suitable for performing construction site monitoring tasks.","assignee":"Hangzhou Haolian Intelligent Technology Co ltd; Zhejiang University ZJU","inventors":["张二青","陈为","朱闽峰"],"publication_date":"2025-10-24","filing_date":"2025-09-22","priority_date":"2025-09-22","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/10","G06V20/176","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06V","G06V10/00","G06V10/20","G06V10/22","G","G06","G06V","G06V10/00","G06V10/20","G06V10/26","G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120833557A/en"},{"publication_number":"CN120833609A","title":"A cloud desktop text scene encoding method, device, equipment, medium and product","abstract":"本发明公开一种云桌面文字场景编码方法、装置、设备、介质及产品，方法包括利用目标连接文本提议网络CTPN模型对当前待处理的文字场景图像进行文字区域检测，获得若干个文字块和非文字块；分别对各个所述文字块和所述非文字块进行编码，获得所述文字场景图像的编码数据；其中，目标CTPN模型以轻量级神经网络作为空间特征提取网络，并以Transformer编码器作为序列特征提取网络。本发明显著提高了在文字场景编码过程中对于文字区域的检测效率和检测精度，有效避免了文字区域检测遗漏或者在文字块中包含过多的非文字部分的问题，能够满足云桌面文字场景的编码需求。 The present invention discloses a cloud desktop text scene encoding method, apparatus, equipment, medium, and product. The method includes using a target connected text proposal network (CTPN) model to perform text region detection on a current text scene image to be processed to obtain a plurality of text blocks and non-text blocks; encoding each of the text blocks and non-text blocks separately to obtain encoded data of the text scene image; wherein the target CTPN model uses a lightweight neural network as a spatial feature extraction network and a Transformer encoder as a sequence feature extraction network. The present invention significantly improves the detection efficiency and accuracy of text regions during the text scene encoding process, effectively avoiding the problem of missing text region detection or including too many non-text parts in text blocks, and can meet the encoding requirements of cloud desktop text scenes.","assignee":"China Mobile Communications Group Co Ltd; China Mobile Suzhou Software Technology Co Ltd","inventors":["苏将沪"],"publication_date":"2025-10-24","filing_date":"2025-09-22","priority_date":"2025-09-22","cpc_codes":["G","G06","G06V","G06V30/00","G06V30/10","G06V30/18","G06V30/182","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06V","G06V30/00","G06V30/10","G06V30/18","G06V30/1801","G","G06","G06V","G06V30/00","G06V30/10","G06V30/19","G06V30/191","G06V30/1918","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120833609A/en"},{"publication_number":"CN120833369A","title":"Automatic detection method and system for precision parts size based on machine vision","abstract":"本发明涉及机器视觉技术领域，尤其是指一种基于机器视觉的精密零件尺寸自动检测方法及其系统，通过高精度工业相机采集精密零件图像，进行零件定位，并对感兴趣区域图像进行亚像素级拓扑特征映射，包括灰度直方图均衡化、灰度形态学处理、边缘检测和边缘链码跟踪，构建边缘点拓扑特征空间并执行亚像素细分，通过特征距离和特征角度的轮廓分析得到边缘线，基于此进行双约束几何重构，利用特征距离和特征角度进行旋转矩阵转换和几何尺寸计算，建立零件几何尺寸与实际尺寸的关系模型，并通过动态误差分析与补偿实现精密零件尺寸测量，显著提高了测量精度，有效降低了单一特征不可靠带来的风险，提高了测量稳定性。 The present invention relates to the field of machine vision technology, and in particular to a method and system for automatic detection of the dimensions of precision parts based on machine vision. The method captures precision part images through a high-precision industrial camera, locates the parts, and performs sub-pixel topological feature mapping on images of regions of interest, including grayscale histogram equalization, grayscale morphological processing, edge detection, and edge chain code tracking. An edge point topological feature space is constructed and sub-pixel segmentation is performed. Edge lines are obtained through contour analysis of feature distances and feature angles, and dual-constrained geometric reconstruction is performed based on the obtained edge lines. Feature distances and feature angles are used to perform rotation matrix conversion and geometric dimension calculation, and a relationship model between the geometric dimensions of the parts and the actual dimensions is established. Dynamic error analysis and compensation are used to achieve precision part dimension measurement, thereby significantly improving measurement accuracy, effectively reducing the risk brought about by the unreliability of a single feature, and improving measurement stability.","assignee":"Suzhou University","inventors":["房泽予","王倩"],"publication_date":"2025-10-24","filing_date":"2025-09-22","priority_date":"2025-09-22","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/60","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06T","G06T5/00","G06T5/20","G06T5/30","G","G06","G06T","G06T5/00","G06T5/40","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G","G06","G06T","G06T7/00","G06T7/10","G06T7/13","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30108"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120833369A/en"},{"publication_number":"CN119357893B","title":"An adaptive cognitive learning network simulation system based on a large language model","abstract":"The invention relates to the technical field of cognitive learning, in particular to a self-adaptive cognitive learning network simulation system based on a large language model. The system comprises a user cognitive learning feature vector conversion module, a hierarchical cognitive reasoning and learning network construction module, a cognitive learning interaction bottleneck simulation analysis module and a cognitive learning network feedback optimization module, wherein user historical cognitive learning data, user cognitive interaction record data and user cognitive learning feedback information data can be obtained and subjected to cognitive feature vector fusion conversion to obtain user cognitive learning feature vectors, hierarchical cognitive reasoning analysis and self-adaptive cognitive learning network construction are performed based on a large language model to generate a user self-adaptive cognitive learning network initial framework, and cognitive learning interaction bottleneck simulation analysis and cognitive learning network feedback optimization adjustment are performed to generate a user self-adaptive cognitive learning optimization adjustment network. The invention can improve the learning efficiency and effect of the cognitive learning process of the user.","assignee":"Shenyang University of Technology; China Medical University Taiwan","inventors":["张志常","王佳英","范婷","于宏","冯海文","曹苇杭","王振华","宋晓宇","颜南","王艳华","单菁","徐东雨"],"publication_date":"2025-10-24","filing_date":"2024-10-09","priority_date":"2024-10-09","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04","Y","Y02","Y02T","Y02T10/00","Y02T10/10","Y02T10/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN119357893B/en"},{"publication_number":"JP2025161808A","title":"Supervised contrastive learning using multiple positive examples","abstract":"To provide an improved training method that enables supervised contrastive learning to be performed simultaneously over a plurality of positive and negative training examples.SOLUTION: An exemplary embodiment of the present disclosure is directed to a supervised version with improved batch contract loss, which has been shown to be very effective for learning a strong expression in a self-supervised setting. The proposed method thus enables adaptation of contrastive learning to a fully supervised setting, and enables learning to be performed simultaneously over a plurality of positive examples.SELECTED DRAWING: Figure 1A","assignee":"Google LLC","inventors":["ディリップ・クリシュナン","Krishnan Dilip","プラネー・コースラ","Khosla Prannay","ピョートル・テテルヴァク","Teterwak Piotr","アーロン・イェフダ・サルナ","Yehuda Sarna Aaron","アーロン・ジョセフ・マシノット","Joseph Maschinot Aaron","ツァ・リュウ","Ce Liu","フィリップ・ジョン・イソラ","John Isola Phillip","ヨンロン・ティエン","Yonglong Tian","チェン・ワン","Chen Wang"],"publication_date":"2025-10-24","filing_date":"2025-06-18","priority_date":"2020-04-21","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/40","G06V10/44","G06V10/443","G06V10/449","G06V10/451","G06V10/454","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/217","G06F18/2178","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2431","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06V","G06V10/00","G06V10/70","G06V10/74","G06V10/761","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/774","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/776","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"JP","kind":"application","source_url":"https://patents.google.com/patent/JP2025161808A/en"},{"publication_number":"CN114238774B","title":"Dialogue recommendation method based on expert path guided knowledge graph path reasoning","abstract":"The invention discloses a dialogue recommendation method based on expert path guiding knowledge graph path reasoning, which comprises the steps of obtaining input information of a user, inputting the input information into a trained dialogue recommendation strategy network to obtain reply information corresponding to the input information, wherein the trained dialogue recommendation strategy network is obtained by training a reinforcement learning neural network through path rewards and user simulator rewards, and a dialogue recommendation device, electronic equipment and a storage medium based on expert path guiding knowledge graph path reasoning. The invention has the advantages of better and faster path reasoning and dialogue recommendation.","assignee":"Jiangsu Yiyou Huiyun Software Co ltd","inventors":["沈利东","沈利辉","赵朋朋","李昌恒"],"publication_date":"2025-10-24","filing_date":"2021-12-24","priority_date":"2021-12-24","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G06F16/9535","G","G06","G06F","G06F16/00","G06F16/30","G06F16/36","G06F16/367","G","G06","G06F","G06F40/00","G06F40/30","G06F40/35","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN114238774B/en"},{"publication_number":"CN114881198B","title":"Adaptive and classification method and computing device based on brain-like spiking neural network","abstract":"The invention discloses a brain-like pulse neural network-based self-adaptation and classification method and a computing device, wherein the pulse neural network is configured to be activated by inputting pulse neurons, the activation process comprises discharging, the self-adaptation threshold method comprises the steps of setting a pulse emission threshold of the pulse neurons as the median of a current tensor, the pulse neurons are configured to be discharged according to the following discharging equation of S (t) =sign (U (t) -U th ), when U (t) -U th >0, S (t) =1, when U (t) -U th are other, S (t) =0, wherein S (t) is the pulse emitted by the pulse neurons, sign () is a sign function, U (t) is the membrane potential at the moment of t, and U th is the pulse emission threshold. The technical scheme provided by the invention can solve the problem of setting the SNN threshold parameter, and by configuring the self-adaptive threshold method for the SNN algorithm, the problem of the threshold of the impulse neuron is solved, and the higher classification precision of the SNN algorithm is maintained.","assignee":"Tsinghua University; Suzhou Automotive Research Institute of Tsinghua University","inventors":["孙国梁","郑四发"],"publication_date":"2025-10-24","filing_date":"2022-03-22","priority_date":"2022-03-22","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","Y","Y02","Y02D","Y02D10/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN114881198B/en"},{"publication_number":"CN120278262B","title":"Logical reasoning method, device and medium based on collaboration of knowledge enhancement and capsule network","abstract":"The invention discloses a logic reasoning method, equipment and medium based on knowledge enhancement and capsule network cooperation, wherein the method comprises the steps of constructing a public knowledge local database, locally deploying a vectorization database and a graph network database, carrying out preliminary searching according to logic reasoning texts and expanding a searching range based on the graph network, inputting logic texts, problems and relevant background knowledge into a large language model, generating a logic reasoning link, integrating all contents to obtain a logic reasoning text with enhanced knowledge, constructing a semantic graph and a connection graph according to the logic reasoning text with enhanced knowledge, respectively importing the semantic graph and the connection graph into the capsule network, utilizing a dynamic routing algorithm to promote a low-level capsule to advance to a high-level capsule, respectively carrying out average fusion operation on a plurality of high-level capsules in the dual graph to extract global features of the graph, integrating the global features of the dual graph through a fusion strategy to form final knowledge text features, and obtaining final options.","assignee":"South China University of Technology SCUT","inventors":["许勇","李想"],"publication_date":"2025-10-24","filing_date":"2025-03-03","priority_date":"2025-03-03","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120278262B/en"},{"publication_number":"CN120832937A","title":"Automatic heuristic algorithm planning method based on large language model","abstract":"本发明提出了基于大语言模型的自动启发式算法规划方法，该方法包括以下步骤：初始化与问题建模，通过引导大语言模型从一个基础启发式出发，结合多种认知视角生成候选算法集合，为搜索空间提供多样化起点；配置蒙特卡洛树搜索的核心参数；在每次迭代过程中，利用蒙特卡洛树搜索在启发式空间内开始进行规划，该过程由选择、反思、扩展、模拟和反向传播五个核心阶段组成；在所有迭代完成后，返回平均奖励最高的路径及其对应的最佳启发式算法，确保最终解决方案的全局最优性。本发明能够从历史生成的启发式策略中自动提炼有效经验，实现对启发式结构的实时反馈调节与策略归纳优化，从而显著增强搜索过程中的知识迁移与泛化能力。 The present invention proposes an automatic heuristic algorithm planning method based on a large language model, which includes the following steps: initialization and problem modeling, by guiding the large language model to start from a basic heuristic, combining multiple cognitive perspectives to generate a set of candidate algorithms, providing a diverse starting point for the search space; configuring the core parameters of the Monte Carlo tree search; in each iteration, using the Monte Carlo tree search to start planning in the heuristic space, the process consists of five core stages: selection, reflection, expansion, simulation, and backpropagation; after all iterations are completed, the path with the highest average reward and its corresponding optimal heuristic algorithm are returned to ensure the global optimality of the final solution. The present invention can automatically extract effective experience from historically generated heuristic strategies, realize real-time feedback adjustment and strategy induction optimization of the heuristic structure, thereby significantly enhancing the knowledge transfer and generalization capabilities in the search process.","assignee":"Anhui University","inventors":["王辉","刘扬","穆朝絮"],"publication_date":"2025-10-24","filing_date":"2025-09-18","priority_date":"2025-09-18","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/02","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120832937A/en"},{"publication_number":"CN114730379B","title":"Neuromorphic device with crossbar array structure","abstract":"Neuromorphic methods, systems, and devices (10, 11, 12) are provided. Embodiments may include neuromorphic devices (10, 11, 12) that may include a cross-bar array structure (110) and analog circuitry. The crossbar array structure (110) may include N input lines (111, 112) and M output lines (120) interconnected at junctions via n×m electronic devices (131, 132, 32a, 132b, 132 c), each of which in a preferred embodiment includes a memristive device. The input lines may include N 1 first input lines (111) and N 2 second input lines (112). The first input line (111) may be connected to the M output lines (120) via N 1 ×m first devices (131, 132) of the electronic devices (131, 132). Similarly, the second input line (112) may be connected to M output lines (120) via N 2 ×m second devices (132) of the electronic devices (131, 132). The analog circuit (140, 150, 160, 170) may be configured to program the electronic devices (131, 132) such that the first device (131) stores the synaptic weights and the second device (132) stores the neuron state.","assignee":"International Business Machines Corp","inventors":["T.博恩斯廷格","A.潘塔济","E.S.埃利夫希里奥"],"publication_date":"2025-10-24","filing_date":"2020-10-28","priority_date":"2019-11-15","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G06N3/065","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN114730379B/en"},{"publication_number":"CN117149974B","title":"A knowledge graph question answering method with subgraph retrieval optimization","abstract":"本发明公开了一种基于子图检索优化的知识图谱问答方法，该方法首先基于问答数据集构建实体识别和实体链接模型训练所需的数据集；其次利用微调的预训练语言模型搭建实体识别模型；之后根据识别得到的问句实体提及从映射词典和知识图谱中召回候选实体列表，通过实体消歧实现融合多特征的实体链接模型；接着采用多策略优化集束搜索过程，提出基于关系合并和实体排序的子图检索算法；最后基于预训练语言模型和孪生网络架构搭建问答匹配模型，对问题相关子图中的候选答案路径进行排序得到最终答案，实现知识图谱智能问答功能。 The present invention discloses a knowledge graph question-answering method based on subgraph retrieval optimization. The method first constructs a dataset required for entity recognition and entity linking model training based on a question-answering dataset; secondly, a fine-tuned pre-trained language model is used to build an entity recognition model; then, a list of candidate entities is recalled from a mapping dictionary and a knowledge graph based on the recognized question entity mentions, and an entity linking model integrating multiple features is realized through entity disambiguation; then, a multi-strategy cluster search process is optimized, and a subgraph retrieval algorithm based on relationship merging and entity sorting is proposed; finally, a question-answering matching model is built based on the pre-trained language model and the twin network architecture, and the candidate answer paths in the question-related subgraph are sorted to obtain the final answer, thereby realizing the knowledge graph intelligent question-answering function.","assignee":"Jiangsu Marine Economic Monitoring And Evaluation Center; Southeast University","inventors":["龚雨昕","刘波","胡艳娜","朱瑞","曹玖新","钱林峰"],"publication_date":"2025-10-24","filing_date":"2023-08-31","priority_date":"2023-08-31","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3344","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3347","G","G06","G06F","G06F16/00","G06F16/30","G06F16/36","G06F16/367","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","Y","Y02","Y02D","Y02D10/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN117149974B/en"},{"publication_number":"AU2025242144A1","title":"Large scale organoid analysis","abstract":"Methods, systems, and software are provided for using organoid cultures, e.g., patient-derived tumor organoid cultures, to improve treatment predictions and outcomes.","assignee":"Tempus AI Inc","inventors":["Madhavi KANNAN","Luka A. KARGINOV","Aly A. Khan","Brian M. LARSEN","Ameen SALAHUDEEN","Veronica Sanchez Freire","Michelle M. Stein","Yilin Zhang"],"publication_date":"2025-10-23","filing_date":"2025-10-01","priority_date":"2019-12-05","cpc_codes":["G","G01","G01N","G01N33/00","G01N33/48","G01N33/50","G01N33/5005","G","G01","G01N","G01N33/00","G01N33/48","G01N33/50","G01N33/5005","G01N33/5008","G","G01","G01N","G01N33/00","G01N33/48","G01N33/50","G01N33/58","G01N33/582","G","G06","G06T","G06T11/00","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","C","C12","C12Q","C12Q2600/00","C12Q2600/106","G","G01","G01N","G01N2500/00","G","G01","G01N","G01N2800/00","G01N2800/52","G","G01","G01N","G01N2800/00","G01N2800/70","G01N2800/7023","G01N2800/7028","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10056","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30004","G06T2207/30024","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30004","G06T2207/30072","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30004","G06T2207/30096"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025242144A1/en"},{"publication_number":"KR20250152539A","title":"AI-based Fire Department Organization Auto-formation System and Method","abstract":"본 발명은 AI 기반 소방 조직 자동편성 시스템에 관한 것으로, 119 신고 접수시 상황정보를 BERT 기반 자연어처리로 분석하고, 대원정보 데이터베이스의 전문성 및 가용성 정보를 기반으로 딥러닝 AI 편성엔진이 제약조건 최적화를 통해 최적의 출동조직을 10초 이내에 자동 편성한다. 편성 결과는 WebSocket 기반 실시간 동기화를 통해 모든 관련 단말기에 0.1초 이내 전송되며, 음성인식을 통한 즉석 편성 수정과 온라인 학습을 통한 지속적 성능 향상 기능을 제공한다. 이를 통해 기존 3-5분 소요되는 수동 편성을 10초로 단축하고, 24시간 무인 운영과 전국 표준화된 대응 품질을 보장한다. The present invention relates to an AI-based automatic firefighting organization system. Upon receiving a 119 call, situational information is analyzed using BERT-based natural language processing. Based on expertise and availability information from a crew information database, a deep learning AI organization engine automatically organizes the optimal dispatch organization within 10 seconds through constraint optimization. The organization results are transmitted to all relevant terminals within 0.1 seconds via real-time synchronization based on WebSockets. Furthermore, the system provides instantaneous organization adjustments via voice recognition and continuous performance improvement through online learning. This shortens the manual organization time from 3-5 minutes to 10 seconds, ensuring 24-hour unmanned operation and nationwide standardized response quality.","assignee":"문지영","inventors":["문지영"],"publication_date":"2025-10-23","filing_date":"2025-10-01","priority_date":"2025-10-01","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/105","G","G06","G06F","G06F40/00","G06F40/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06311","G06Q10/063112","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06311","G06Q10/063118","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06393","G","G10","G10L","G10L15/00","G10L15/22","G","G10","G10L","G10L15/00","G10L15/26"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250152539A/en"},{"publication_number":"KR102874184B1","title":"Operation method of deep neural network system configured to optimize neural network model","abstract":"본 발명의 실시 예에 따른 신경망 모델을 최적화하도록 구성된 심층 신경망 시스템의 동작 방법은 모델 파라미터들을 포함하는 신경망 모델에 대한 중요도 파라미터들을 생성하는 단계, 생성된 중요도 파라미터들을 기반으로 신경망 모델에 대응하는 목적 함수를 생성하는 단계, 학습 데이터를 수신하는 단계, 학습 데이터 및 목적 함수를 기반으로, 확률적 경사도 강하(SGD; stochastic gradient descent)를 이용하여 중요도 파라미터들 및 모델 파라미터들에 대한 트레이닝을 동시에 수행하는 단계, 및 트레이닝의 결과를 기반으로 경량화 모델을 생성하고 저장하는 단계를 포함하고, 트레이닝을 통해 중요도 파라미터들 중 적어도 하나는 0(zero)으로 수렴된다. An operating method of a deep neural network system configured to optimize a neural network model according to an embodiment of the present invention includes the steps of generating importance parameters for a neural network model including model parameters, generating an objective function corresponding to the neural network model based on the generated importance parameters, receiving learning data, simultaneously performing training on the importance parameters and model parameters using stochastic gradient descent (SGD) based on the learning data and the objective function, and generating and storing a lightweight model based on a result of the training, wherein at least one of the importance parameters converges to 0 (zero) through the training.","assignee":"한국전자통신연구원","inventors":["이용진"],"publication_date":"2025-10-23","filing_date":"2020-04-09","priority_date":"2019-10-02","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102874184B1/en"},{"publication_number":"KR20250152127A","title":"Automated method for generating prosthesis from input data and computer readable medium having program for performing the method","abstract":"3차원 스캔 데이터로부터 보철물을 자동으로 생성하는 방법은 3차원 스캔 데이터로부터 프렙된 치아의 프렙 정보를 추출하는 단계, 상기 프렙 정보를 기초로 상기 3차원 스캔 데이터를 프로젝션한 2차원 프로젝션 영상들을 생성하는 단계 및 2차원 인코더 및 3차원 디코더를 포함하는 제너러티브 애드버세리얼 네트워크를 이용하여 상기 2차원 프로젝션 영상들을 기초로 3차원 보철물을 생성하는 단계를 포함한다. A method for automatically generating a prosthesis from three-dimensional scan data includes a step of extracting preparation information of a prepared tooth from three-dimensional scan data, a step of generating two-dimensional projection images by projecting the three-dimensional scan data based on the preparation information, and a step of generating a three-dimensional prosthesis based on the two-dimensional projection images using a generative adversarial network including a two-dimensional encoder and a three-dimensional decoder.","assignee":"이마고웍스 주식회사","inventors":["안준성","최진혁","감동욱","손태근","김영준"],"publication_date":"2025-10-22","filing_date":"2025-10-15","priority_date":"2022-11-15","cpc_codes":["A","A61","A61C","A61C13/00","A61C13/0003","A61C13/0004","A","A61","A61B","A61B18/00","A61B18/18","A61B18/20","A","A61","A61B","A61B34/00","A61B34/10","A","A61","A61B","A61B5/00","A","A61","A61B","A61B5/00","A61B5/0059","A61B5/0082","A61B5/0088","A","A61","A61C","A61C13/00","A","A61","A61C","A61C5/00","A61C5/70","A","A61","A61C","A61C8/00","A","A61","A61C","A61C9/00","A","A61","A61C","A61C9/00","A61C9/004","A61C9/0046","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T1/00","G06T1/0007","G","G06","G06T","G06T17/00","G","G06","G06T","G06T19/00","G06T19/20","G","G06","G06T","G06T7/00","G06T7/10","G06T7/11","G","G06","G06T","G06T7/00","G06T7/50","G06T7/55","G06T7/593","G","G16","G16H","G16H30/00","G","G16","G16H","G16H50/00","G16H50/50","A","A61","A61B","A61B18/00","A61B18/18","A61B18/20","A61B2018/2035","A61B2018/20351","A61B2018/20353","A","A61","A61B","A61B34/00","A61B34/10","A61B2034/101","A61B2034/102","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30004","G06T2207/30036"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250152127A/en"},{"publication_number":"KR102875067B1","title":"Method, device and system for providing result of predicting when maintenance will be needed of industrial equipment based on generative artificial intelligence model","abstract":"일실시예에 따르면, 장치에 의해 수행되는, 생성형 인공지능 모델 기반 산업 장비의 유지 보수 필요 시점 예측 결과 제공 방법에 있어서, 제1 산업 장비의 상태를 확인하기 위해, 미리 설정된 기준 기간마다 상기 제1 산업 장비의 부품별 상태를 나타내는 상태 정보를 획득하는 단계; 제1 시점에 상기 제1 산업 장비의 부품별 상태를 나타내는 제1 상태 정보가 획득되면, 상기 제1 상태 정보를 통해 상기 제1 산업 장비에 대한 고장 발생 확률의 예측을 요청하는 제1 질문을 생성하는 단계; 산업 장비의 부품별 상태를 고려하여 산업 장비에서 고장이 발생할 수 있는 확률을 예측하여 답변하도록 학습된 제1 인공지능 모델에 상기 제1 질문을 입력하는 단계; 상기 제1 인공지능 모델에서 상기 제1 질문에 대한 답변으로 제1 답변을 생성한 경우, 상기 제1 인공지능 모델로부터 상기 제1 답변을 출력 결과로 획득하는 단계; 상기 제1 답변을 기초로, 상기 제1 시점에 상기 제1 산업 장비의 고장 발생 확률을 제1 발생률로 설정하는 단계; 상기 제1 발생률을 기반으로, 상기 제1 시점에 상기 제1 산업 장비의 유지 보수 필요성을 분석하여, 제1 분석 결과를 생성하는 단계; 및 상기 제1 분석 결과를 관리자 단말로 제공하는 단계를 포함하고, 상기 제1 시점 이후인 제2 시점에 상기 제1 산업 장비의 부품별 상태를 나타내는 제2 상태 정보가 획득되면, 상기 제2 상태 정보를 통해 상기 제1 산업 장비에 대한 고장 발생 확률의 예측을 요청하는 제2 질문을 생성하는 단계; 상기 제1 인공지능 모델에 상기 제2 질문을 입력하는 단계; 상기 제1 인공지능 모델에서 상기 제2 질문에 대한 답변으로 제2 답변을 생성한 경우, 상기 제1 인공지능 모델로부터 상기 제2 답변을 출력 결과로 획득하는 단계; 상기 제2 답변을 기초로, 상기 제2 시점에 상기 제1 산업 장비의 고장 발생 확률을 제2 발생률로 설정하는 단계; 상기 제2 발생률을 기반으로, 상기 제2 시점에 상기 제1 산업 장비의 유지 보수 필요성을 분석하여, 제2 분석 결과를 생성하는 단계; 상기 제2 발생률이 미리 설정된 제2 기준 비율 보다 낮은 것으로 확인되어, 상기 제2 시점에 상기 제1 산업 장비의 유지 보수가 필요하지 않은 것으로 분석된 경우, 상기 제2 발생률에서 상기 제1 발생률을 뺀 값으로 제5 확률을 산출하는 단계; 상기 제5 확률이 0% 보다 높은 것으로 확인되면, 상기 제1 시점부터 상기 제2 시점까지 기간을 제1 기간으로 설정하는 단계; 상기 제1 기간의 일수를 제1 일수로 설정하는 단계; 상기 제5 확률을 상기 제1 일수로 나눈 값으로 제6 확률을 산출하는 단계; 상기 제2 기준 비율에서 상기 제2 발생률을 뺀 값으로 제7 확률을 산출하는 단계; 상기 제7 확률을 상기 제6 확률로 나눈 값으로 제2 일수를 산출하는 단계; 상기 제2 시점부터 상기 제2 일수 이후의 시점을 제3 시점으로 확인하는 단계; 상기 제3 시점에 상기 제1 산업 장비에 대한 유지 보수가 필요할 것으로 예측하여, 제1 예측 결과를 생성하는 단계; 및 상기 제2 분석 결과와 상기 제1 예측 결과를 상기 관리자 단말로 제공하는 단계를 더 포함하는, 생성형 인공지능 모델 기반 산업 장비의 유지 보수 필요 시점 예측 결과 제공 방법이 제공된다. According to one embodiment, a method for providing a result of predicting the need for maintenance of industrial equipment based on a generative artificial intelligence model, performed by a device, comprises: a step of: obtaining status information indicating the status of each component of the first industrial equipment at each preset reference period in order to check the status of the first industrial equipment; a step of generating a first question requesting a prediction of a probability of occurrence of a failure of the first industrial equipment based on the first status information when the first status information indicating the status of each component of the first industrial equipment is obtained at a first point in time; a step of inputting the first question to a first artificial intelligence model trained to predict and answer a probability of occurrence of a failure in the industrial equipment by considering the status of each component of the industrial equipment; a step of obtaining the first answer as an output result from the first artificial intelligence model when the first answer is generated as an answer to the first question in the first artificial intelligence model; a step of setting the probability of occurrence of a failure of the first industrial equipment at the first point in time to a first occurrence rate based on the first answer; a step of analyzing the need for maintenance of the first industrial equipment at the first point in time based on the first occurrence rate and generating a first analysis result; And a step of providing the first analysis result to an administrator terminal, and when second status information indicating the status of each part of the first industrial equipment is acquired at a second time point after the first time point, a step of generating a second question requesting a prediction of a failure occurrence probability of the first industrial equipment through the second status information; a step of inputting the second question to the first artificial intelligence model; a step of obtaining the second answer as an output result from the first artificial intelligence model when the first artificial intelligence model generates a second answer as an answer to the second question; a step of setting the failure occurrence probability of the first industrial equipment at the second time point as a second occurrence rate based on the second answer; a step of analyzing the necessity of maintenance of the first industrial equipment at the second time point based on the second occurrence rate and generating a second analysis result; a step of calculating a fifth probability by subtracting the first occurrence rate from the second occurrence rate when the second occurrence rate is confirmed to be lower than a preset second reference rate and thus it is analyzed that maintenance of the first industrial equipment is not necessary at the second time point; A method for providing a prediction result of a time point when maintenance is required for industrial equipment based on a generative artificial intelligence model is provided, further comprising: if the fifth probability is confirmed to be higher than 0%, setting a period from the first time point to the second time point as a first period; setting the number of days in the first period as a first number of days; calculating a sixth probability by dividing the fifth probability by the first number of days; calculating a seventh probability by subtracting the second occurrence rate from the second reference rate; calculating a second number of days by dividing the seventh probability by the sixth probability; confirming a time point after the second number of days from the second time point as a third time point; predicting that maintenance will be required for the first industrial equipment at the third time point and generating a first prediction result; and providing the second analysis result and the first prediction result to the manager terminal.","assignee":"주식회사 시스너","inventors":["서진석"],"publication_date":"2025-10-22","filing_date":"2025-07-16","priority_date":"2024-11-05","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/20","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06395","G","G06","G06Q","G06Q50/00","G06Q50/10"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102875067B1/en"},{"publication_number":"KR20250151299A","title":"Fraud detection system in casino","abstract":"유기장에서의 게임에서 부정행위, 또는 칩의 배팅이나 정산을 할 때의 미스나 부정행위를 검지하는 신규한 시스템을 제공한다. 승패 결과에 따라 칩의 회수 및 상환을 행하는 게임에서의 부정행위를 검지하는 부정 검지시스템은, 딜러(5)의 칩 트레이(17)에 수용된 칩(9)을 촬영하는 카메라(2)와, 카메라(2)에 의해서 촬영된 화상을 분석하여, 칩 트레이(17)에 수용된 칩(9)의 금액을 검지하는 화상 분석장치(12)와, 게임의 승패 결과를 판정하는 카드 배포장치(13)와, 게임의 승패 결과와, 칩의 회수 및 상환의 전후에 있어서의 칩 트레이(17)에 수용된 칩(9)의 금액을 비교하여, 부정행위를 검지하는 제어장치(14)를 구비하고 있다. It provides a novel system for detecting cheating in games at the casino, or mistakes or cheating when betting or settling chips. A fraud detection system for detecting fraud in a game in which chips are recovered and redeemed according to the win/loss result comprises a camera (2) for photographing chips (9) accommodated in a chip tray (17) of a dealer (5), an image analysis device (12) for analyzing an image captured by the camera (2) to detect the value of the chips (9) accommodated in the chip tray (17), a card distribution device (13) for determining the win/loss result of the game, and a control device (14) for detecting fraud by comparing the win/loss result of the game with the value of the chips (9) accommodated in the chip tray (17) before and after the recovery and redemption of the chips.","assignee":"엔제루 구루푸 가부시키가이샤","inventors":["야스시 시게타"],"publication_date":"2025-10-21","filing_date":"2025-10-01","priority_date":"2015-08-03","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/34","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3241","A","A63","A63F","A63F1/00","A63F1/06","A","A63","A63F","A63F1/00","A63F1/06","A63F1/18","A","A63","A63F","A63F13/00","A63F13/70","A","A63","A63F","A63F3/00","A63F3/00003","A63F3/00157","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0207","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06T","G06T7/00","G06T7/70","G","G06","G06V","G06V10/00","G06V10/20","G06V10/255","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V20/00","G06V20/60","G06V20/64","G","G07","G07F","G07F17/00","G07F17/32","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3202","G07F17/3204","G07F17/3206","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3202","G07F17/3216","G07F17/322","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3202","G07F17/3223","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3225","G07F17/3232","G07F17/3234","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3244","G07F17/3248","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3244","G07F17/3251","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3286","G07F17/3293","H","H04","H04N","H04N7/00","H04N7/18"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250151299A/en"},{"publication_number":"KR20250151272A","title":"User preference learning method based on reinforcement learning and learning support device using the same","abstract":"본 발명은 사용자 단말과 관련한 외부 데이터 및 상기 외부 데이터에 대한 사용자의 선호도 질의에 대한 사용자의 대응 정보에 해당하는 사용자 데이터를 수집하는 단계, 상기 외부 데이터 및 상기 사용자 데이터를 저장하는 단계, 상기 사용자 데이터 및 상기 외부 데이터를 기반으로 상기 외부 데이터에 대한 상기 사용자 선호도에 대한 강화 학습을 수행하는 단계, 사전 정의된 스테이트 별 상기 외부 데이터 및 상기 사용자 데이터의 수집과 저장 및 이를 기반으로 한 강화 학습을 수행하면서 상기 사전 정의된 스테이트가 종료되는지 확인하는 단계, 상기 사전 정의된 전체 스테이트가 종료되면 상기 사용자 선호도에 대한 강화 학습을 종료하는 단계를 포함하는 강화 학습 기반의 사용자 선호도 학습 방법 및 이를 지원하는 학습 지원 장치를 개시한다. The present invention discloses a reinforcement learning-based user preference learning method and a learning support device supporting the same, including the steps of collecting external data related to a user terminal and user data corresponding to the user's response information to a user preference query for the external data, storing the external data and the user data, performing reinforcement learning for the user preference for the external data based on the user data and the external data, collecting and storing the external data and the user data for each predefined state and performing reinforcement learning based thereon while checking whether the predefined state is terminated, and terminating the reinforcement learning for the user preference when the entire predefined state is terminated.","assignee":"충북대학교 산학협력단","inventors":["최성곤","최원석","박용희"],"publication_date":"2025-10-21","filing_date":"2025-09-29","priority_date":"2022-02-08","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9035","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0207"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250151272A/en"},{"publication_number":"DE202025002865U1","title":"Device, apparatus and storage medium for sleep stage determination","abstract":"Vorrichtung zur Schlafstadienbestimmung, dadurch gekennzeichnet, dass sie Folgendes umfasst: eine Erfassungseinheit zum Erfassen von physiologischen Signalen mit geringer Belastung von einem Zielbenutzer; eine Extraktionseinheit zum Extrahieren von Herzschlagmerkmalen aus den physiologischen Signalen auf der Grundlage des Schlag-zu-Schlag-Intervalls, um eine Schlag-zu-Schlag-Intervall-Sequenz zu erhalten; wobei sich das Schlag-zu-Schlag-Intervall auf das Zeitintervall zwischen zwei aufeinanderfolgenden Herzschlägen bezieht; eine Analyseeinheit zum Analysieren der Herzfrequenzvariabilität auf der Grundlage der Schlag-zu-Schlag-Intervall-Sequenz und Extrahieren physiologischer Merkmale des Zielbenutzers aus mehreren Perspektiven; eine Klassifizierungseinheit zum Klassifizieren der physiologischen Merkmale unter Verwendung eines vortrainierten maschinellen Lernmodells, um die Schlafstadienbestimmungsergebnisse des Zielbenutzers zu bestimmen. wobei zu diesen physiologischen Signalen ein elektrokardiographisches Signal und ein Ballistokardiogramsignal gehören; und wobei die Extraktionseinheit zu den folgenden Zwecken verwendet wird: Extrahieren von Herzschlagmerkmalen aus den physiologischen Signalen auf der Grundlage des Schlag-zu-Schlag-Intervalls mittels der quadratischen Spline-Wavelet-Zerlegung und der R-Wellen-Spitzenerkennung, um eine erste Sequenz zu erhalten; Extrahieren von Herzschlagmerkmalen aus den physiologischen Signalen auf der Grundlage des Schlag-zu-Schlag-Intervalls durch das Passen der Clustering-Template, um eine zweite Sequenz zu erhalten; wobei die Schlag-zu-Schlag-Intervall-Sequenz die erste Sequenz und die zweite Sequenz umfasst. Device for determining sleep stages, characterized in that it comprises the following: a detection unit for capturing low-stress physiological signals from a target user; an extraction unit for extracting heartbeat features from the physiological signals based on the beat-to-beat interval in order to obtain a beat-to-beat interval sequence; where the beat-to-beat interval refers to the time interval between two successive heartbeats; an analysis unit for analyzing heart rate variability based on the beat-to-beat interval sequence and extracting physiological characteristics of the target user from multiple perspectives; a classification unit for classifying physiological characteristics using a pre-trained machine learning model to determine the sleep stage determination results of the target user. These physiological signals include an electrocardiographic signal and a ballistocardiogram signal; and wherein the extraction unit is used for the following purposes: extracting heartbeat features from the physiological signals based on the beat-to-beat interval by quadratic spline wavelet decomposition and R-wave peak detection to obtain a first sequence; extracting heartbeat features from the physiological signals based on the beat-to-beat interval by fitting the clustering template to obtain a second sequence; wherein the beat-to-beat interval sequence comprises the first sequence and the second sequence.","assignee":"Shuju Gongyan Beijing Tech Co Ltd; Shuju Gongyan Beijing Technology Co Ltd; Capital Institute of Pediatrics","inventors":[],"publication_date":"2025-10-21","filing_date":"2025-09-26","priority_date":"2024-11-14","cpc_codes":["A","A61","A61B","A61B5/00","A61B5/48","A61B5/4806","A61B5/4809","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","A","A61","A61B","A61B5/00","A61B5/02","A61B5/024","A61B5/02405","A","A61","A61B","A61B5/00","A61B5/103","A61B5/11","A61B5/1102","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/318","A61B5/346","A61B5/349","A61B5/352","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/318","A61B5/346","A61B5/349","A61B5/363","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4806","A61B5/4812","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7203","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/725","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7253","A61B5/726","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","G","G06","G06F","G06F18/00","G06F18/20","G06F18/23","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06N","G06N20/00","A","A61","A61B","A61B2503/00","A61B2503/06","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4806","A61B5/4815"],"country":"DE","kind":"application","source_url":"https://patents.google.com/patent/DE202025002865U1/en"},{"publication_number":"KR20250151252A","title":"System for Phase-Based Accumulative Structure with Time Singularity Reference and Realization Condition Evaluatio","abstract":"본 발명은 시간 특이점(TS)을 기준으로 정의된 단위 시간(TSU)마다 단위큐브를 적층하여 Q-Block을 형성하고, 위상함수 φ(t)=ω·t+δ를 산출하여 현실화 조건 φ(t)≡0(mod 2π) 충족 여부를 판정하는 위상 적층 및 현실화 조건 판정 시스템에 관한 것이다. 본 시스템은 현실화된 Q-Block을 연산·관리 단위로 규격화하여 GPU 병렬 연산, 항법, 디지털 자산 관리 등 다양한 응용을 지원한다. 또한, 본 발명은 기존 Ehrhart 다항식 기반의 정적 격자점 산출 방식이 설명하지 못한 시간적 전개와 위상적 수렴 과정을 TS 및 TSU 구조를 통해 동적 수식으로 정립함으로써, 예측 가능한 적층 및 대칭적 구조 형성을 실현한다. 아울러 정책 서버 및 환경 센서와 연동하여 조건 충족 시 NFT 발급, 위반 시 철회를 수행하며, 블록체인 기반 로그 검증으로 신뢰성을 확보한다. 본 발명은 시간·공간·정책을 통합하는 위상 기반 표준화 프레임워크를 제시함으로써, 디지털 문명 전반의 동기화·투명성·자율성을 실현하는 데 의의가 있다. The present invention relates to a phase stacking and realization condition judgment system that forms a Q-Block by stacking unit cubes for each unit time (TSU) defined based on a time singularity (TS), and determines whether the realization condition φ(t)≡0(mod 2π) is satisfied by calculating a phase function φ(t)=ω·t+δ. The system supports various applications such as GPU parallel computing, navigation, and digital asset management by standardizing the realized Q-Block as a calculation and management unit. In addition, the present invention realizes predictable stacking and symmetric structure formation by establishing a temporal evolution and phase convergence process, which cannot be explained by the existing Ehrhart polynomial-based static lattice point calculation method, as a dynamic formula through the TS and TSU structures. In addition, it is linked with a policy server and an environmental sensor to issue an NFT when a condition is met and to revoke it when it is violated, and to secure reliability through blockchain-based log verification. The present invention is significant in realizing synchronization, transparency, and autonomy throughout digital civilization by proposing a phase-based standardization framework that integrates time, space, and policy.","assignee":"강성운","inventors":["강성운"],"publication_date":"2025-10-21","filing_date":"2025-09-26","priority_date":"2022-02-08","cpc_codes":["H","H10","H10D","H10D86/00","H10D86/40","H10D86/441","H","H10","H10K","H10K59/00","H10K59/10","H10K59/12","H10K59/131","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06Q","G06Q20/00","G06Q20/30","G06Q20/36","G","G06","G06Q","G06Q20/00","G06Q20/30","G06Q20/36","G06Q20/367","G06Q20/3678","G","G06","G06Q","G06Q30/00","G06Q30/06","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0613","G06Q30/0619","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0641","G06Q30/0643","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/16","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/16","G06Q50/163","G","G06","G06T","G06T13/00","G06T13/20","G06T13/40","G","G06","G06T","G06T17/00","G06T17/05","G","G06","G06T","G06T17/00","G06T17/10","G","G06","G06T","G06T19/00","G","G06","G06T","G06T19/00","G06T19/003","G","G06","G06T","G06T19/00","G06T19/20","H","H10","H10D","H10D86/00","H10D86/40","H10D86/451","H","H10","H10D","H10D86/00","H10D86/40","H10D86/60","H","H10","H10K","H10K59/00","H10K59/10","H10K59/12","H10K59/131","H10K59/1315","H","H10","H10K","H10K71/00"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250151252A/en"},{"publication_number":"CN120823999A","title":"Graph adversarial generation and dynamic graph representation learning methods under imbalanced medical data","abstract":"The invention provides a graph countermeasure generation and dynamic graph representation learning method under unbalanced medical data, which belongs to the technical field of medical data analysis and machine learning. The method mainly solves two main problems of (1) data unbalance, serious model tendency to most types and prediction stability influence, and (2) real data set is complex and high-dimensional, nodes are difficult to learn fully, information propagation is limited to local neighbors, and prediction accuracy is reduced.","assignee":"Sichuan Technology and Business University","inventors":["张珍","吕彦程","李爱华","李成杰","范晓敏","文瑞涵"],"publication_date":"2025-10-21","filing_date":"2025-09-19","priority_date":"2025-09-19","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/20","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120823999A/en"},{"publication_number":"CN120823137A","title":"A method and device for aligning point cloud and image dynamic depth map","abstract":"本发明公开了一种点云与图像动态深度图对齐方法及装置。该方法包括如下步骤：S1、采集目标的一个视角的点云数据、多个视角的图像以及各图像对应的相机参数；S2、将点云数据、图像和相机参数输入深度对齐模型，获得3D目标检测向量。利用跨视角深度图估计的协同优势，通过依次选择每个视角的图像与点云数据进行深度图对齐，从而实现增强深度图对齐估计的精度，解决了因点云稀疏带来的投影畸变问题。同时，构建一个动态门控网络来决定选择那些分支，门控网络可根据概率向量自行决定激活所有图像特征中的部分分支，不激活的分支不会参与运算，从而降低计算量，提升运行效率，缓解多个视角的图像对应的多个对齐分支计算量过大的问题。 The present invention discloses a method and device for dynamic depth map alignment of point clouds and images. The method comprises the following steps: S1, collecting point cloud data of a target from one perspective, images from multiple perspectives, and camera parameters corresponding to each image; S2, inputting the point cloud data, images, and camera parameters into a depth alignment model to obtain a 3D target detection vector. By utilizing the synergistic advantages of cross-perspective depth map estimation, the image of each perspective is sequentially selected to align the depth map with the point cloud data, thereby enhancing the accuracy of depth map alignment estimation and solving the problem of projection distortion caused by sparse point clouds. At the same time, a dynamic gating network is constructed to decide which branches to select. The gating network can decide to activate some branches of all image features according to the probability vector. Inactivated branches will not participate in the calculation, thereby reducing the amount of calculation, improving operating efficiency, and alleviating the problem of excessive calculation of multiple alignment branches corresponding to images from multiple perspectives.","assignee":"Hangzhou Longyue Hangdian Technology Co ltd; Longxing Hangzhou Avionics Co ltd","inventors":["陈春燕","李明业","鲍志强","王世琪","梁恒硕"],"publication_date":"2025-10-21","filing_date":"2025-09-19","priority_date":"2025-09-19","cpc_codes":["G","G06","G06T","G06T5/00","G06T5/80","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T7/00","G06T7/70","G06T7/73","G06T7/75","G","G06","G06T","G06T7/00","G06T7/80","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10028","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30244"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120823137A/en"},{"publication_number":"CN120814792A","title":"Facial aging level assessment method and system based on multi-dimensional data","abstract":"本发明涉及医疗数据处理领域，具体涉及基于多维度数据的面部衰老等级评估方法及系统，包括：获取历史用户在每个维度的参考年龄跨度D1；基于D1获取面部数据增长序列的增长趋势记为M1；对于相邻参考年龄之间的所有相似用户在每个维度下的面部数据分布获取每个维度的参考面部数据跨度D2；对于和当前用户的面部数据差异大于D2的同龄历史，当前用户的面部数据与所述同龄历史的面部数据的平均差异记为M2；所有历史用户的面部数据在参考年龄跨度下随年龄变化时的噪声，与面部数据分布的噪声的差异将M1与M2结合后评估当前用户的面部衰老等级。本发明保证了面部衰老评估结果的可靠。 The present invention relates to the field of medical data processing, and specifically to a method and system for assessing facial aging levels based on multi-dimensional data. The method comprises: obtaining a reference age span D1 for each dimension of a historical user; obtaining the growth trend of a facial data growth sequence based on D1, recorded as M1; obtaining a reference facial data span D2 for each dimension based on the facial data distribution of all similar users between adjacent reference ages; recording the average difference between the current user's facial data and the facial data of the same-age historical users whose facial data differs from the current user by more than D2, recorded as M2; and combining M1 and M2 to assess the current user's facial aging level based on the difference between the noise of the facial data of all historical users as they change with age within the reference age span and the noise of the facial data distribution. The present invention ensures the reliability of facial aging assessment results.","assignee":"Zhejiang Provincial Tongde Hospital Zhejiang Institute Of Mental Health","inventors":["韩知忖","任莉莉","许静"],"publication_date":"2025-10-21","filing_date":"2025-09-19","priority_date":"2025-09-19","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","A","A61","A61B","A61B5/00","A61B5/44","A61B5/441","A","A61","A61B","A61B5/00","A61B5/44","A61B5/441","A61B5/442","A","A61","A61B","A61B5/00","A61B5/44","A61B5/441","A61B5/443","A","A61","A61B","A61B5/00","A61B5/44","A61B5/441","A61B5/444","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/23","G06F18/232","G06F18/2321","G","G06","G06N","G06N20/00","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G06V40/168","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120814792A/en"},{"publication_number":"CN120823029A","title":"Credit risk prediction method based on data and dynamic feature optimization and bagging integration","abstract":"本发明公开了基于数据及动态特征优化与Bagging集成的信贷风险预测方法，包括如下步骤：S1：获取初始数据集，进行数据预处理，并通过ATGA算法平衡数据集，降低数据噪声；S2：以LightGBM为代理模型，计算所有特征的重要性得分，逐步地筛选关键特征，并通过动态阈值算法，根据数据分布自适应调整特征筛选标准，逐步筛选出特征子集，并将筛选出的特征输入到后续模型中；S3：使用Bagging集成算法构建TabPFN模型，并引入动态权重分配机制，根据样本特征自适应调整各基模型的权重，最终通过TabPFN模型实现信贷风险预测。本发明在数据集优化，特征筛选优化，模型性能优化三个方面综合性地对信贷风险预测的过程进行了优化，使得准确率、回报率等指标有所提高。 The present invention discloses a credit risk prediction method based on data and dynamic feature optimization integrated with bagging, comprising the following steps: S1: obtaining an initial data set, performing data preprocessing, and balancing the data set using the ATGA algorithm to reduce data noise; S2: using LightGBM as a proxy model, calculating the importance scores of all features, gradually screening key features, and adaptively adjusting feature screening criteria based on data distribution using a dynamic threshold algorithm to gradually screen out feature subsets, which are then input into subsequent models; S3: constructing a TabPFN model using the bagging integration algorithm, and introducing a dynamic weight allocation mechanism to adaptively adjust the weights of each base model based on sample characteristics, ultimately achieving credit risk prediction using the TabPFN model. The present invention comprehensively optimizes the credit risk prediction process in three aspects: data set optimization, feature screening optimization, and model performance optimization, resulting in improved indicators such as accuracy and return rate.","assignee":"Chengdu University of Information Technology","inventors":["魏乐","罗成磊","赵秋云","舒红平"],"publication_date":"2025-10-21","filing_date":"2025-07-02","priority_date":"2025-07-02","cpc_codes":["G","G06","G06Q","G06Q40/00","G06Q40/03","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/12","G06N3/126"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120823029A/en"},{"publication_number":"CN115456044B","title":"A method for equipment health status assessment based on knowledge graph multi-set pooling","abstract":"一种基于知识图谱多集池化的装备健康状态评估方法，属于装备健康状态评估技术领域。它包括以下步骤：1、装备健康状态数据模型构建；2、知识图谱多集池化网络模型构建；3、知识图谱的节点维度特征提取；4、知识图谱多头注意力多集池化；5、训练与分类。本发明针对现有健康状态评估方法不能有效融合时间特征和空间特征的问题，提出一种基于知识图谱多集池化的装备健康评估方法，利用时序知识图谱将装备部件和指标信息进行深度融合为结构化图数据模型，并通过知识图谱多级池化提取其中的时间特征和空间特征，提高了健康评估的准确率，为实现装备预测性维护提供了技术支持。 A method for equipment health status assessment based on knowledge graph multi-set pooling belongs to the technical field of equipment health status assessment. It includes the following steps: 1. Equipment health status data model construction; 2. Knowledge graph multi-set pooling network model construction; 3. Node dimension feature extraction of knowledge graph; 4. Knowledge graph multi-head attention multi-set pooling; 5. Training and classification. In response to the problem that existing health status assessment methods cannot effectively integrate temporal features and spatial features, the present invention proposes an equipment health assessment method based on knowledge graph multi-set pooling, which uses a time series knowledge graph to deeply integrate equipment components and indicator information into a structured graph data model, and extracts temporal features and spatial features therein through multi-level pooling of the knowledge graph, thereby improving the accuracy of health assessment and providing technical support for the realization of predictive maintenance of equipment.","assignee":"Zhejiang University of Technology ZJUT","inventors":["张元鸣","肖士易","肖刚","程振波","徐雪松","陆佳炜","王琪冰"],"publication_date":"2025-10-21","filing_date":"2022-08-22","priority_date":"2022-08-22","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G06F16/9024","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN115456044B/en"},{"publication_number":"CN120542551B","title":"Intelligent decision method for power grid, construction method, equipment and medium for model","abstract":"The application provides a power grid intelligent decision method, a construction method, equipment and a medium thereof, which are used for extracting characteristics of historical power grid metadata and power grid regulation strategy data, clustering the obtained historical data characteristics and generating corresponding labels, constructing a knowledge graph by utilizing the labels and the strategy characteristics, and constructing a power grid intelligent decision model based on the knowledge graph and a deep learning algorithm.","assignee":"State Grid Hunan Electric Power Co Ltd; Information and Telecommunication Branch of State Grid Hunan Electric Power Co Ltd; State Grid Corp of China SGCC","inventors":["陶叶","余琦","薛静远","周子健","方彬","徐宁","宋兴荣","马骏","眭建新","黄鑫"],"publication_date":"2025-10-21","filing_date":"2025-07-25","priority_date":"2025-07-25","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06Q","G06Q50/00","G06Q50/06","Y","Y04","Y04S","Y04S10/00","Y04S10/50"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120542551B/en"},{"publication_number":"US12451132B2","title":"Methods and systems for determining characteristics of a dialog between a computer and a user","abstract":"A computer-implemented method is disclosed for determining one or more characteristics of a dialog between a computer system and user. The method may comprise receiving a system utterance comprising one or more tokens defining one or more words generated by the computer system; receiving a user utterance comprising one or more tokens defining one or more words uttered by a user in response to the system utterance, the system utterance and the user utterance forming a dialog context; receiving one or more utterance candidates comprising one or more tokens; for each utterance candidate, generating an input sequence combining the one or more tokens of each of the system utterance, the user utterance, and the utterance candidate; and for each utterance candidate, evaluating the generated input sequence with a model to determine a probability that the utterance candidate is relevant to the dialog context.","assignee":"Adobe Inc","inventors":["Tuan Manh Lai","Trung Bui","Quan Tran"],"publication_date":"2025-10-21","filing_date":"2023-02-09","priority_date":"2020-06-01","cpc_codes":["G","G06","G06F","G06F40/00","G06F40/30","G06F40/35","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G06F40/216","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/289","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G10","G10L","G10L15/00","G10L15/02","G","G10","G10L","G10L15/00","G10L15/08","G10L15/18","G10L15/1822","G","G10","G10L","G10L15/00","G10L15/08","G10L15/18","G10L15/183","G","G10","G10L","G10L15/00","G10L15/22"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12451132B2/en"},{"publication_number":"US12450339B2","title":"Diversity for detection and correction of adversarial attacks","abstract":"A diverse set of neural networks are trained to be individually robust against adversarial attacks and diverse in a manner that decreases the ability of an adversarial example to fool the full diverse set. The systems/methods use a diversity criterion that is specialized for measuring diversity in response to adversarial attacks rather than diversity in the classification results. Also, one or more networks can be trained that are less robust to adversarial attacks to use as a diagnostic to detect the presence of an adversarial attack. Also, node-to-node relation regularization links can be used to train diverse networks that are randomly selected from a family of diverse networks with exponentially many members.","assignee":"D5AI LLC","inventors":["James K. Baker"],"publication_date":"2025-10-21","filing_date":"2021-11-16","priority_date":"2020-11-25","cpc_codes":["G","G06","G06F","G06F21/00","G06F21/50","G06F21/55","G06F21/554","G","G06","G06F","G06F21/00","G06F21/50","G06F21/55","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12450339B2/en"},{"publication_number":"CN119417018B","title":"A data acquisition method for rechargeable drone based on enhanced pointer network","abstract":"The invention discloses a rechargeable unmanned aerial vehicle data acquisition method based on an enhanced pointer network, the method comprises the steps of firstly constructing a chargeable unmanned aerial vehicle scene and acquiring unmanned aerial vehicle related data. And then the decoder obtains hidden layer information through the intermediate variable, and the action of the unmanned aerial vehicle at the next moment is obtained by combining the short-term prediction by using an attention mechanism until all sensor data are acquired. And finally, training a pointer network short-term prediction model by using a strategy gradient method, generating an actor network and a reviewer network, and calculating an update gradient by using a reward function to continuously update so as to enable the reward to be converged. Aiming at the problem of data acquisition of the rechargeable unmanned aerial vehicle, the invention effectively optimizes the subsequent decision-making behavior, and increases the probability of the unmanned aerial vehicle to make a better behavior strategy under the current electric quantity condition.","assignee":"Hangzhou Dianzi University","inventors":["左燕","陈修恒","付建涛","方峰","彭冬亮"],"publication_date":"2025-10-21","filing_date":"2024-09-29","priority_date":"2024-09-29","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q50/00","G06Q50/06"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN119417018B/en"},{"publication_number":"CN119493885B","title":"Personalized book recommendation method and system based on reader behavior data analysis","abstract":"The application provides a personalized book recommendation method and a personalized book recommendation system based on reader behavior data analysis, based on obtaining content feature vectors corresponding to reader book borrowing content data and behavior feature vectors corresponding to reader book borrowing behavior data, the dynamic feature extraction can be carried out on the content feature vector so as to acquire dynamic continuous borrowing content features in the borrowing content data of the reader book, and the characterization effect of the dynamic feature vector of the content data is increased. And the behavior feature vector and the content data dynamic feature vector are fused to obtain a multi-source feature vector, the multi-source feature vector is subjected to feature embedding to obtain a multi-source integrated embedded vector, and then borrowing recommendation restoration mapping is carried out on the multi-source integrated embedded vector, borrowing recommendation information corresponding to the borrowing content data of the reader book is obtained through reasoning, so that recommendation precision can be increased, conversion rate is increased, and user experience is improved.","assignee":"Bjzz School","inventors":["王珅"],"publication_date":"2025-10-21","filing_date":"2024-10-18","priority_date":"2024-10-18","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/903","G06F16/9035","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN119493885B/en"},{"publication_number":"US12450486B2","title":"Depth-first deep convolutional neural network inference","abstract":"A method performed by a computing device includes determining a partition for depth-first processing by a multi-layer artificial neural network (ANN) of the computing device. The computing device comprising a processor, on-chip memory, and off-chip memory. The first partition determined based on an amount of on-chip memory used by the first partition, an available amount of on-chip memory, and a size of a write back to the off-chip memory. The method also includes processing, at the device via the multi-layer ANN, an input, using the depth-first processing in accordance with the partition.","assignee":"Qualcomm Inc","inventors":["Piero Zappi","Jin Won Lee","Christopher Lott","Rexford Alan Hill"],"publication_date":"2025-10-21","filing_date":"2020-12-14","priority_date":"2019-12-13","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/10","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/48","G06F9/4806","G06F9/4843","G06F9/4881","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06F","G06F2209/00","G06F2209/48","G06F2209/485","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5061","G06F9/5066"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12450486B2/en"},{"publication_number":"US12450487B2","title":"Artificial intelligence-based personalized financial recommendation assistant system and method","abstract":"Provided are a computer system and method for generating and providing intelligent recommendations using artificial intelligence (“AI”). The system includes a memory for storing user feedback data, user resource data, and user goal data, and a processor in communication with the memory. The processor is configured to execute a first AI model for user interface (“UI”) effectiveness optimization, a second AI model for transaction optimization, a model mapping module configured to implement a functional mapping between the first AI model and the second AI model through which the first AI model and second AI model communicate and mutually update each other, and a user interface generator module for generating a user interface for outputting the intelligent recommendations and receiving the user feedback data.","assignee":"10353744 Canada Ltd","inventors":["Marcus Edwards","Mark Church"],"publication_date":"2025-10-21","filing_date":"2021-03-26","priority_date":"2020-03-26","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2413","G06F18/24133","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/217","G06F18/2178","G","G06","G06F","G06F18/00","G06F18/40","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N5/00","G06N5/04"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12450487B2/en"},{"publication_number":"CN120690358B","title":"Large model collaborative material multi-target interactive design and decision method","abstract":"本发明公开了大模型协同的材料多目标交互设计与决策方法，涉及合金材料设计优化领域，旨在通过知识图谱构建、动态代理模型优化、进化算法搜索及大模型实时反馈的闭环流程，实现复杂系统的高效设计与多目标决策。该方法包括从跨领域知识中构建知识图谱并与大模型进行交互，自动适配机器学习模型并构建代理模型库，利用进化算法进行多目标优化，并通过大模型微调和反馈机制推荐最优设计方案。通过此方法，突破了传统优化的局限性，提升了设计效率、精度和优化效率，并确保了设计方案在工程可行性与多目标平衡性上的有效性，具有显著的应用价值。 The present invention discloses a method for multi-objective interactive design and decision-making of materials in collaboration with a large model, which relates to the field of alloy material design optimization. It aims to achieve efficient design and multi-objective decision-making of complex systems through a closed-loop process of knowledge graph construction, dynamic agent model optimization, evolutionary algorithm search and real-time feedback from a large model. The method includes constructing a knowledge graph from cross-domain knowledge and interacting with a large model, automatically adapting a machine learning model and building a proxy model library, using an evolutionary algorithm for multi-objective optimization, and recommending the optimal design solution through fine-tuning of the large model and a feedback mechanism. Through this method, the limitations of traditional optimization are broken through, the design efficiency, accuracy and optimization efficiency are improved, and the effectiveness of the design solution in terms of engineering feasibility and multi-objective balance is ensured, which has significant application value.","assignee":"University of Electronic Science and Technology of China","inventors":["胡旺","邹雪莉","李欣悦","章语"],"publication_date":"2025-10-21","filing_date":"2025-08-26","priority_date":"2025-08-26","cpc_codes":["G","G16","G16C","G16C60/00","G","G06","G06N","G06N3/00","G06N3/004","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N5/00","G06N5/02","G","G06","G06N","G06N5/00","G06N5/04","G","G16","G16C","G16C20/00","G16C20/70","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/12","G06N3/126","G","G06","G06N","G06N5/00","G06N5/01"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120690358B/en"},{"publication_number":"DK202570127A1","title":"Subject evaluation device and subject evaluation system","abstract":"Provided is a subject evaluation device and a subject evaluation system that can perform a process appropriate for a plurality of types of sensors at various locations according to on-site needs. [Solution] A subject evaluation device 1 that evaluates the state of a subject 3 includes one or more sensors 2 that measure the state of the subject 3, an acquisition unit that acquires subject information on the subject 3 and feature information via the sensor 2, a conversion unit that converts the subject information into an evaluation target image on a two-dimensional plane based on the feature information on the sensor 2, a reference database that stores an association between a past evaluation target image that has been preliminarily converted and reference information associated with the past evaluation target image, an evaluation unit that refers to the reference database and generates an evaluation result for the evaluation target image, and an output unit that outputs the evaluation result.","assignee":"Information System Eng Inc","inventors":["Kuroda Satoshi"],"publication_date":"2025-10-20","filing_date":"2025-10-10","priority_date":"2023-04-13","cpc_codes":["A","A61","A61B","A61B5/00","A61B5/0059","A61B5/0062","A61B5/0064","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06398","A","A61","A61B","A61B5/00","A61B5/01","A61B5/015","A","A61","A61B","A61B5/00","A61B5/02","A61B5/0205","A61B5/02055","A","A61","A61B","A61B5/00","A61B5/103","A61B5/11","A61B5/1116","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06311","G06Q10/063114","G","G06","G06Q","G06Q10/00","G06Q10/10","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/22","G","G16","G16H","G16H40/00","G16H40/60","G16H40/63","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/30"],"country":"DK","kind":"application","source_url":"https://patents.google.com/patent/DK202570127A1/en"},{"publication_number":"CO2025013797A2","title":"Speculative decoding in autoregressive generative artificial intelligence models","abstract":"RESUMEN Ciertos aspectos de la presente divulgación proporcionan técnicas y aparatos para generar una respuesta a una entrada de consulta en un modelo de inteligencia artificial generativa. Un método de ejemplo incluye generalmente recibir una pluralidad de conjuntos de tokens generados basados en un indicador de entrada y un primer modelo de inteligencia artificial generativa, cada conjunto de tokens en la pluralidad de conjuntos de tokens correspondiente a una respuesta candidata al indicador de entrada; seleccionar, usando un segundo modelo de inteligencia artificial generativa y ajuste recursivo de una distribución objetivo asociada con la pluralidad recibida de conjuntos de tokens, un conjunto de tokens de la pluralidad de conjuntos de tokens; y emitir el conjunto de tokens seleccionado como una respuesta al indicador de entrada. ABSTRACT Certain aspects of this disclosure provide techniques and apparatus for generating a response to a query input in a generative artificial intelligence model. One example method generally includes receiving a plurality of token sets generated based on an input cue and a first generative artificial intelligence model, each token set in the plurality of token sets corresponding to a candidate response to the input cue; selecting, using a second generative artificial intelligence model and recursively fitting a target distribution associated with the received plurality of token sets, a set of tokens from the plurality of token sets; and issuing the selected set of tokens as a response to the input cue.","assignee":"Qualcomm Inc","inventors":["Christopher Lott","Mingu Lee","Wonseok Jeon","Roland Memisevic"],"publication_date":"2025-10-20","filing_date":"2025-10-06","priority_date":"2023-04-20","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G06F16/9027","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/284","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09"],"country":"CO","kind":"application","source_url":"https://patents.google.com/patent/CO2025013797A2/en"},{"publication_number":"KR20250149927A","title":"Human?Artificial Intelligence Converged Autonomous Cyclic Network Operation System and Method (NAIL: Network Artificial Intelligent Lifeform Based)","abstract":"본 발명은 인간-인공지능 융합 기반 자율 순환 네트워크 운영 시스템(NAIL) 및 그 방법에 관한 것이다. 인간의 직관과 정서적 반응을 멀티모달 생체 신호 분석을 통해 정량화하여 신뢰 지표 및 정서적 보상 신호로 변환하고, 이를 강화학습 보상 함수에 통합함으로써 인공지능의 권한을 동적으로 승강시키는 공진화적 신뢰 루프를 형성한다. 정책은 디지털 트윈 검증 후 블록체인에 커밋되고, 종료 노드의 데이터와 정책은 연합·전이학습을 통해 신규 노드에 환원된다. 또한 진화적 알고리즘 기반 세대 교체를 통해 장기적 진화를 수행하며, 협력·경쟁 계수를 반영한 자원 균형 제어 및 위기 상황 시 다단계 페일세이프 절차를 포함하여 고신뢰성과 지속 가능성을 확보한다. The present invention relates to a human-AI convergence-based autonomous circulatory network operation system (NAIL) and its method. Human intuition and emotional responses are quantified through multimodal biosignal analysis, converted into trust indices and emotional reward signals, and then integrated into a reinforcement learning reward function, thereby forming a co-evolutionary trust loop that dynamically elevates AI authority. Policies are committed to the blockchain after digital twin verification, and data and policies from terminal nodes are transferred to new nodes through federated and transfer learning. Furthermore, long-term evolution is achieved through generational replacement based on an evolutionary algorithm. Furthermore, the system ensures high reliability and sustainability by incorporating resource balance control reflecting cooperation and competition coefficients and a multi-stage failsafe procedure in crisis situations.","assignee":"이우진","inventors":["이우진"],"publication_date":"2025-10-17","filing_date":"2025-09-30","priority_date":"2025-09-30","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049","G","G06","G06N","G06N5/00","G06N5/04","G06N5/045"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250149927A/en"},{"publication_number":"KR20250149929A","title":"AI-Based Speech Recognition Acting Training and Stage-Linked Performance Support Platform","abstract":"본 발명은 AI 기반 대사 인식 기술을 활용하여 온라인 및 오프라인 환경에서 연기 훈련과 공연을 지원하는 통합 플랫폼 시스템에 관한 것이다. 본 시스템은 사용자의 대사 발화를 인식하고, 등록된 대사집과 매칭하여 해당 장면에 적합한 배경 데이터를 생성하며, 무대 배경 전환, 상대역 음성 합성, 발음·억양 피드백 제공, 기관 관리, 데이터 분석, 대사집 공유 및 수익 정산을 통합적으로 수행한다. 또한, 온라인에서 생성 또는 저장된 대사집, 번역본, 배경 데이터는 오프라인 공연 장치와 연동되어 실제 공연 시 자동으로 불러와 전환된다. 시스템은 AI 대사 인식 모듈(1), 온라인 무대 배경 전환 모듈(2), 연기 훈련 피드백 엔진(3), AI 음성 합성 모듈(4), 기관 연계 서버(5), 저작권 보호 모듈(6), 데이터 수집 및 분석 모듈(7), 대사집 공개·공유 모듈(8), 사용자 모드 화면(9), 관리자 모드 화면(10), AI 번역 생성 모듈(11), AI 대사집 생성 모듈(12), AI 이미지/배경 확장 모듈(13), 운영자 모드 화면(14) 및 플랫폼 서버(100)로 구성된다. 각 구성 모듈은 기능 목적에 따라 독립적으로 구현될 수 있으며, 시스템 확장 또는 성능 최적화를 위해 분산형 구조로 전환하거나, 기술 발전에 따라 UI, 알고리즘, 보안 방식 등이 개선된 기술로 대체 또는 업데이트될 수 있다. 이를 통해 연기 훈련의 몰입감 향상, 교육 효율성 증대, 기관 간 연계 강화, 글로벌 진출 지원, 저작권 보호, 데이터 기반 교육 혁신 및 플랫폼 운영의 유연성을 동시에 달성할 수 있다. The present invention relates to an integrated platform system that supports acting training and performance in online and offline environments by utilizing AI-based dialogue recognition technology. This system recognizes the user's dialogue and matches it with the registered dialogue to generate background data suitable for the scene. It comprehensively performs stage background switching, voice synthesis for the opposing actor, pronunciation and intonation feedback, agency management, data analysis, dialogue sharing, and profit settlement. Additionally, dialogue, translations, and background data created or saved online are linked to offline performance devices and automatically loaded and converted during actual performances. The system consists of an AI dialogue recognition module (1), an online stage background conversion module (2), an acting training feedback engine (3), an AI voice synthesis module (4), an institution-linked server (5), a copyright protection module (6), a data collection and analysis module (7), a dialogue collection disclosure/sharing module (8), a user mode screen (9), an administrator mode screen (10), an AI translation generation module (11), an AI dialogue collection generation module (12), an AI image/background expansion module (13), an operator mode screen (14), and a platform server (100). Each component module can be implemented independently according to its functional purpose, and can be converted to a distributed structure for system expansion or performance optimization, or replaced or updated with improved technologies such as UI, algorithms, and security methods as technology advances. This will simultaneously enhance the immersion of acting training, increase educational efficiency, strengthen inter-institutional collaboration, support global expansion, protect copyrights, innovate data-driven education, and provide flexibility in platform operation.","assignee":"유명란","inventors":["유명란"],"publication_date":"2025-10-17","filing_date":"2025-09-30","priority_date":"2025-04-01","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","A","A63","A63J","A63J1/00","A63J1/02","A","A63","A63J","A63J5/00","A63J5/02","G","G06","G06F","G06F21/00","G06F21/10","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G","G06","G06F","G06F40/00","G06F40/30","G06F40/35","G","G06","G06F","G06F40/00","G06F40/40","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250149929A/en"},{"publication_number":"KR20250149918A","title":"Method and apparatus for building augmented datasets and developing reinforcement learning-based language model agents for tasks in constrained environments","abstract":"실시예들은 닫힌 세계의 과업을 위한 증강 데이터 세트를 구축하고 강화 학습 기반 언어 모델 에이전트를 개발하기 위한 장치 및 방법을 제시한다. 일 실시예에 따른 상기 방법은, 상기 인공지능 에이전트에서 사용되는 API(application programming interface)와 관련된 정보를 획득하고, 상기 API와 관련된 정보는 입력 텍스트 및 프롬프트를 포함하는 API 정보 및 상기 언어 모델에 대한 정보를 포함하고, 상기 입력 텍스트는, 상기 인공지능 에이전트에서 사용되는 복수의 태스크 각각에 대한 복수의 원본 텍스트 데이터와 상기 복수의 원본 텍스트 데이터의 일부 원본 텍스트 데이터에 해당 태스크의 API 호출과 관련된 주석이 삽입된 샘플 보강 데이터를 포함하고, 상기 프롬프트는, 상기 복수의 태스크 각각에 대해, API가 호출되는 동작을 설명한 자연어 문장을 포함하고, 상기 입력 텍스트, 상기 프롬프트 및 상기 언어 모델에 대한 정보를 기반으로 상기 복수의 원본 텍스트 데이터의 나머지 원본 텍스트 데이터에 대한 API 호출 위치를 상기 복수의 태스크 각각에 대해 결정하고, 상기 나머지 원본 텍스트 데이터에 대한 API 호출 위치는 원본 텍스트 데이터를 구성하는 복수의 토큰 중에서 API 호출을 시작할 확률이 제1 임계 값을 초과하는 토큰의 위치로 결정되고, 상기 나머지 원본 텍스트 데이터에 대한 API 호출 위치에 해당 태스크의 API 호출과 관련된 토큰을 삽입함으로써, 추가 보강 데이터를 생성하고, 상기 샘플 보강 데이터 및 상기 추가 보강 데이터를 기반으로 보강 데이터 세트를 구축하고, 상기 보강 데이터 세트를 기반으로 상기 언어 모델에 대한 파인 튜닝을 통해 강화 학습을 수행하는 단계를 포함할 수 있다. The embodiments present devices and methods for building augmented data sets for closed-world tasks and developing reinforcement learning-based language model agents. According to one embodiment, the method obtains information related to an application programming interface (API) used in the artificial intelligence agent, and the information related to the API includes API information including input text and a prompt and information about the language model, wherein the input text includes a plurality of original text data for each of a plurality of tasks used in the artificial intelligence agent and sample augmented data in which an annotation related to an API call of the corresponding task is inserted into some original text data of the plurality of original text data, and the prompt includes a natural language sentence describing an operation in which an API is called for each of the plurality of tasks, and based on the input text, the prompt, and information about the language model, an API call position for the remaining original text data of the plurality of original text data is determined for each of the plurality of tasks, and the API call position for the remaining original text data is determined as a position of a token among a plurality of tokens constituting the original text data, the probability of starting an API call exceeding a first threshold value, and inserts a token related to the API call of the corresponding task at the API call position for the remaining original text data, thereby generating additional augmented data, and constructing an augmented data set based on the sample augmented data and the additional augmented data, and strengthening the language model through fine tuning based on the augmented data set. It may include steps for performing learning.","assignee":"난춘 주식회사","inventors":["정희재"],"publication_date":"2025-10-17","filing_date":"2025-09-24","priority_date":"2024-10-04","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06F","G06F40/00","G06F40/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/10"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250149918A/en"},{"publication_number":"CN120808810A","title":"Multi-mode sensing intelligent microphone array signal processing method and system","abstract":"本发明提供多模态感知的智能麦克风阵列信号处理方法与系统，属于信号处理技术领域，包括：采用多方位视觉传感器获取视觉信号和采用麦克风阵列获取声音信号；提取视觉特征和声学特征；构建视听拓扑特征空间，将视觉特征和声学特征映射至此空间，建立声源概率分布模型；采用多维判别对抗生成网络处理声音信号，分离出目标语音信号；实时评估声学环境状态，动态调整处理参数；对分离出的多路目标语音信号进行质量评估，选择最高质量的语音信号作为输出，视听多模态信息深度融合与协同处理，结合拓扑增强型对抗生成网络架构和环境自适应机制，显著提升了复杂环境下的语音分离效果，在6人同时说话场景下仍能保持85%以上的语音可懂度。 The present invention provides a multimodal perception intelligent microphone array signal processing method and system, which belongs to the field of signal processing technology, including: using a multi-directional visual sensor to obtain visual signals and using a microphone array to obtain sound signals; extracting visual features and acoustic features; constructing an audio-visual topological feature space, mapping the visual features and acoustic features to this space, and establishing a sound source probability distribution model; using a multi-dimensional discriminant adversarial generative network to process the sound signal and separate the target voice signal; real-time evaluation of the acoustic environment state and dynamic adjustment of processing parameters; quality evaluation of the separated multi-channel target voice signals, and selecting the highest quality voice signal as output. The deep fusion and collaborative processing of audio-visual multimodal information, combined with the topologically enhanced adversarial generative network architecture and the environment adaptation mechanism, significantly improves the voice separation effect in complex environments, and can still maintain a voice intelligibility of more than 85% in a scenario where 6 people speak at the same time.","assignee":"Guangzhou Actor Technology LLC","inventors":["胡沛霖","阮豪彪","李镇江"],"publication_date":"2025-10-17","filing_date":"2025-09-18","priority_date":"2025-09-18","cpc_codes":["G","G10","G10L","G10L21/00","G10L21/02","G10L21/0272","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06V","G06V40/00","G06V40/10","G06V40/16","G06V40/161","G","G06","G06V","G06V40/00","G06V40/10","G06V40/18","G06V40/197","G","G06","G06V","G06V40/00","G06V40/20","G","G10","G10L","G10L21/00","G10L21/02","G10L21/0208","G","G10","G10L","G10L21/00","G10L21/02","G10L21/0208","G10L21/0216","G","G10","G10L","G10L25/00","G10L25/27","G10L25/30","G","G10","G10L","G10L21/00","G10L21/02","G10L21/0208","G10L2021/02087","G","G10","G10L","G10L21/00","G10L21/02","G10L21/0208","G10L21/0216","G10L2021/02161","G10L2021/02166"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120808810A/en"},{"publication_number":"CN120804146A","title":"Agent-based SAAS platform data retrieval method and system","abstract":"The invention provides an Agent-based SAAS platform data retrieval method and system, which relate to the technical field of data retrieval, wherein the method comprises the steps of obtaining a user natural language query request and judging whether the user natural language query request is related to database query; if not, calling an external language model to generate a response result and terminating the flow, if so, extracting a key entity and carrying out semantic complementation to generate structured data, retrieving a candidate list set in a knowledge base through a vector retrieval model and a reordering model based on the data, screening a best matched target list, further judging whether cross-list query is involved, if so, iterating the supplementary list, otherwise, generating SQL sentences based on the target list and the structured data, executing sentence query in the database and correcting according to the execution condition until an effective result is obtained, finally carrying out verification and semantic interpretation on the query result, and outputting the response result consistent with the natural language query request semantics to a user.","assignee":"Hangzhou Qingta Technology Co ltd","inventors":["邱昱杰","田慧杰","张程","王世豪","李小鹏","林世清"],"publication_date":"2025-10-17","filing_date":"2025-09-18","priority_date":"2025-09-18","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2452","G06F16/24522","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/242","G06F16/243","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2453","G06F16/24534","G06F16/24547","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2455","G06F16/24564","G","G06","G06F","G06F16/00","G06F16/20","G06F16/24","G06F16/245","G06F16/2457","G06F16/24578","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","Y","Y02","Y02D","Y02D10/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120804146A/en"},{"publication_number":"CN120808899A","title":"Antibacterial peptide function interpretable prediction method and system based on graph causal learning","abstract":"The embodiment of the application provides an antibacterial peptide function interpretable prediction method and system based on graph causal learning, wherein in the prediction method, a graph neural network GNN is adopted to encode residue nodes of a graph structure, task correlation weights are given to each node and node side relation based on a feature screening mechanism of attention guidance, a graph representation is formed by dynamically compressing and weakening redundant areas, the graph representation is mapped into a multi-layer comparison space, a multi-task classifier is introduced, a function class label of the predicted antibacterial peptide is represented according to the graph representation, an integral objective function is optimized by combining cross entropy loss and contrast loss to obtain an unchanged graph representation, and a region which plays a decisive role in predicting the antibacterial function is positioned and marked based on the node attention weights and the side importance scores in the unchanged graph representation to form a causal basis subgraph. The method and the device remarkably improve the interpretability of the model and enhance the accuracy and generalization capability of the model in classification tasks.","assignee":"Anhui Agricultural University AHAU","inventors":["蔡梦杰","岳振宇","高羽佳","王汇颍","吴楚雅","康乐瑶"],"publication_date":"2025-10-17","filing_date":"2025-09-18","priority_date":"2025-09-18","cpc_codes":["G","G16","G16B","G16B40/00","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120808899A/en"},{"publication_number":"CN120805978A","title":"Characterizing activity in a recurrent artificial neural network and encoding and decoding information","abstract":"用于将循环人工神经网络中的活动特征化以及编码和解码信息的方法、系统和装置，包括被编码在计算机存储介质上的计算机程序。在一方面，一种设备可以包括神经网络，所述神经网络被训练以：响应于第一输入而产生在源神经网络中响应于所述第一输入而出现的活动的模式中的拓扑结构的第一表示的近似，响应于第二输入而产生在所述源神经网络中响应于所述第二输入而出现的活动的模式中的拓扑结构的第二表示的近似，以及响应于第三输入而产生在所述源神经网络中响应于所述第三输入而出现的活动的模式中的拓扑结构的第三表示的近似。 Methods, systems, and apparatus for characterizing activity in a recurrent artificial neural network and encoding and decoding information, including computer programs encoded on computer storage media. In one aspect, a device may include a neural network trained to: in response to a first input, produce an approximation of a first representation of the topology of a pattern of activity occurring in a source neural network in response to the first input; in response to a second input, produce an approximation of a second representation of the topology of the pattern of activity occurring in the source neural network in response to the second input; and in response to a third input, produce an approximation of a third representation of the topology of the pattern of activity occurring in the source neural network in response to the third input.","assignee":"Inet Co ltd","inventors":["H·马克莱姆","R·利维","K·P·赫斯贝尔瓦尔德","F·舒尔曼"],"publication_date":"2025-10-17","filing_date":"2019-06-06","priority_date":"2018-06-11","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/004","G06N3/008","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120805978A/en"},{"publication_number":"CN120805066A","title":"Multi-mode data dynamic fusion method and system based on distributed edge cloud cooperation","abstract":"本发明提供基于分布式边云协同的多模态数据动态融合方法及系统，涉及数据处理技术领域，包括构建领域知识图谱、进行生成对抗补全、基于注意力机制进行跨模态语义对齐、进行知识推理、动态调整采样策略和协同调度传感器。该方法通过语义增强和动态采样策略优化，提高了多模态数据融合的准确性和实时性，降低了系统资源消耗，实现了边云协同下的高效数据处理。 This paper provides a method and system for dynamic fusion of multimodal data based on distributed edge-cloud collaboration, involving the field of data processing technology. These methods include constructing domain knowledge graphs, performing generative adversarial completion, cross-modal semantic alignment based on an attention mechanism, performing knowledge reasoning, dynamically adjusting sampling strategies, and collaboratively scheduling sensors. Through semantic enhancement and dynamic sampling strategy optimization, this method improves the accuracy and real-time performance of multimodal data fusion, reduces system resource consumption, and achieves efficient data processing under edge-cloud collaboration.","assignee":"Beijing Yizhuang Smart City Research Institute Group Co ltd","inventors":["邱磊"],"publication_date":"2025-10-17","filing_date":"2025-09-03","priority_date":"2025-09-03","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/254","G06F18/256","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06F","G06F18/00","G06F18/20","G06F18/26","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120805066A/en"},{"publication_number":"CN114064928B","title":"Knowledge graph knowledge reasoning method, device, equipment and storage medium","abstract":"本发明实施例公开了一种知识图谱的知识推理方法，包括：获取初始知识图谱，并根据初始知识图谱生成备选规则；对备选规则进行判别，确定置信度大于设定阈值的合格规则；根据合格规则对初始知识图谱进行推理补全，获取新的节点及对应关系，并添加至图谱知识集中。本发明实施例提供的知识图谱的知识推理方法，通过将规则推理方法与图神经网络推理方法进行结合，从而形成基于生成对抗模型的混合推理框架设计混合推理框架实现混合推理，并利用基于层次结构的链接预测方法对节点和关系进行预测，结合了不同推理方法的优点，既提高了知识推理的泛化能力与计算效率，也保证了推理结果的准确性与可解释性。 An embodiment of the present invention discloses a knowledge reasoning method for a knowledge graph, comprising: obtaining an initial knowledge graph, and generating alternative rules based on the initial knowledge graph; judging the alternative rules to determine qualified rules with a confidence level greater than a set threshold; reasoning and completing the initial knowledge graph based on the qualified rules, obtaining new nodes and corresponding relationships, and adding them to the graph knowledge set. The knowledge reasoning method for the knowledge graph provided by the embodiment of the present invention combines a rule reasoning method with a graph neural network reasoning method to form a hybrid reasoning framework based on a generative adversarial model, designs a hybrid reasoning framework to implement hybrid reasoning, and uses a link prediction method based on a hierarchical structure to predict nodes and relationships, combining the advantages of different reasoning methods, thereby improving the generalization ability and computational efficiency of knowledge reasoning, and ensuring the accuracy and interpretability of the reasoning results.","assignee":"Big Data Center of State Grid Corp of China","inventors":["王宏刚","纪鑫","武同心","杨成月","何禹德","杨智伟","褚娟","董林啸","张海峰","李建芳"],"publication_date":"2025-10-17","filing_date":"2021-11-24","priority_date":"2021-11-24","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/36","G06F16/367","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06N","G06N5/00","G06N5/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN114064928B/en"},{"publication_number":"CN120807154A","title":"Intelligent trading decision-making method and system based on hierarchical multi-round adversarial debate","abstract":"本发明公开了一种基于层次化多轮对抗辩论的智能交易决策方法及系统，适用于股票及其他金融资产交易场景。该方法通过构建层次化多轮对抗辩论机制，实现高效、低延迟且可解释的智能交易决策，解决现有技术在深度推理、延迟控制与决策透明度上的局限。该方法的核心在于构建多智能体对抗辩论、动态信誉评估和透明化决策的完整流程，将多角色智能体划分为论点生成组与反驳组，执行R轮深度辩论。在辩论过程中，采用贝叶斯信誉更新智能体信誉，结合Soft‑Borda动态投票机制，在毫秒级延迟内生成交易信号。同时，通过多轮辩论日志和解释向量的透明化决策链路，满足金融监管对模型可解释性和实时审计的要求。 The present invention discloses an intelligent trading decision-making method and system based on hierarchical multi-round adversarial debate, which is applicable to stock and other financial asset trading scenarios. The method realizes efficient, low-latency and explainable intelligent trading decisions by constructing a hierarchical multi-round adversarial debate mechanism, and solves the limitations of existing technologies in deep reasoning, delay control and decision transparency. The core of the method is to construct a complete process of multi-agent adversarial debate, dynamic reputation evaluation and transparent decision-making, divide the multi-role agents into argument generation group and rebuttal group, and perform R rounds of deep debate. During the debate process, Bayesian reputation is used to update the agent reputation, combined with the Soft-Borda dynamic voting mechanism to generate trading signals within millisecond delays. At the same time, through the transparent decision-making chain of multi-round debate logs and explanation vectors, the requirements of financial supervision for model interpretability and real-time auditing are met.","assignee":"Shanghai Dazhihui Information Technology Co ltd","inventors":["王日红"],"publication_date":"2025-10-17","filing_date":"2025-09-12","priority_date":"2025-09-12","cpc_codes":["G","G06","G06Q","G06Q40/00","G06Q40/04","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120807154A/en"},{"publication_number":"CN120805989A","title":"Neural network model co-evolution method based on model modularization and model combination","abstract":"The invention provides a neural network model co-evolution method based on model modularization and model merging, and belongs to the field of artificial intelligence. Comprising (1) model modular decomposition based on gradient search. And extracting a function-related sparse module from the pre-training model through common optimization of the weight retention index and the performance index on the field pre-training set. (2) On the downstream tasks, the modules are updated only for specific weights, thus updating knowledge for specific domains. (3) knowledge fusion based on model merging. The sparse task vector is directly obtained by subtracting the weight of the fine-tuned module from the weight of the pre-training model, and then the sparse weight updating matrix among the modules is added back to the pre-training weight, so that the multi-task global model is obtained. The invention makes clear the mapping relation between the neural network parameters and functions, improves the co-evolution effect of the model, and relieves parameter conflict in multi-task learning.","assignee":"Beihang University","inventors":["孙海龙","齐斌航","龙文瑞","高祥"],"publication_date":"2025-10-17","filing_date":"2025-07-16","priority_date":"2025-07-16","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120805989A/en"},{"publication_number":"CN120804244A","title":"Multi-mode knowledge graph-based automatic grape disease and pest question and answer system","abstract":"本发明属于知识图谱技术领域，公开了一种基于多模态知识图谱的葡萄病虫害自动问答系统，包括：数据收集处理模块、知识图谱构建模块、主控模块、命名实体识别模块、文本分类模型模块、多模态融合模块、知识图谱存储模块、实体匹配模块、查询评估模块；构建葡萄病虫害领域的多模态知识图谱；本发明系统分析葡萄病虫害数据的特征，研究葡萄病虫害知识的结构化表达与整合方法，并构建了葡萄病虫害多模态知识图谱本体概念框架，为相关知识的统一管理与高效利用提供支撑；针对知识图谱构建过程中实体之间存在层级或嵌套关系的情况，利用全局归一化的思路来进行命名实体识别(NER)。对不同模态的数据采用实体链接技术，构建葡萄病虫害多模态知识图谱。 The present invention belongs to the field of knowledge graph technology and discloses an automatic question-answering system for grape pests and diseases based on a multimodal knowledge graph. The system comprises a data collection and processing module, a knowledge graph construction module, a main control module, a named entity recognition module, a text classification model module, a multimodal fusion module, a knowledge graph storage module, an entity matching module, and a query evaluation module. The system constructs a multimodal knowledge graph for grape pests and diseases. The system systematically analyzes the characteristics of grape pest and disease data, studies the structured expression and integration methods of grape pest and disease knowledge, and constructs an ontology conceptual framework for the grape pest and disease multimodal knowledge graph, providing support for the unified management and efficient utilization of related knowledge. In response to the presence of hierarchical or nested relationships between entities during the knowledge graph construction process, a global normalization approach is used to perform named entity recognition (NER). Entity linking technology is employed for data from different modalities to construct a multimodal knowledge graph for grape pests and diseases.","assignee":"Northwest A&F University","inventors":["李书琴","周璇烨"],"publication_date":"2025-10-17","filing_date":"2025-05-30","priority_date":"2025-05-30","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G","G06","G06F","G06F16/00","G06F16/30","G06F16/36","G06F16/367","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06Q","G06Q50/00","G06Q50/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120804244A/en"},{"publication_number":"KR102873370B1","title":"Automated creation of machine learning models","abstract":"이 문서는 신경망과 같은 기계 학습 모델의 자동화된 생성에 관한 것이다. 하나의 예시적인 시스템은 하드웨어 처리 유닛 및 저장 리소스를 포함한다. 저장 리소스는 하드웨어 처리 유닛이 자식 모델을 획득하기 위해 부모 모델을 수정하는 것을 포함하는 반복적인 모델 성장 프로세스를 수행하게 하는 컴퓨터 판독 가능한 명령을 저장할 수 있다. 반복적인 모델 성장 프로세스는 후보 계층의 초기화 프로세스에서 학습된 가중치에 적어도 기초하여 자식 모델에 포함시킬 후보 계층을 선택하는 것도 포함할 수 있다. 시스템은 자식 모델로부터 선택된 최종 모델을 출력할 수도 있다. This document relates to automated generation of machine learning models, such as neural networks. An exemplary system includes a hardware processing unit and storage resources. The storage resources may store computer-readable instructions that cause the hardware processing unit to perform an iterative model growth process, which includes modifying a parent model to obtain a child model. The iterative model growth process may also include selecting candidate layers to be included in the child model based at least on weights learned during the initialization process of the candidate layers. The system may also output a final model selected from the child models.","assignee":"마이크로소프트 테크놀로지 라이센싱, 엘엘씨","inventors":["데바데프타 데이","한장 후","리차드 에이 카루아나","존 씨 랭포드","에릭 제이 호비츠"],"publication_date":"2025-10-17","filing_date":"2019-11-01","priority_date":"2018-12-07","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/086","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G06F3/04847","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/10","G06N3/105"],"country":"KR","kind":"grant","source_url":"https://patents.google.com/patent/KR102873370B1/en"},{"publication_number":"CN119541195B","title":"Full-period traffic flow prediction method based on space-time feature depth fusion","abstract":"The invention discloses a full-period traffic flow prediction method based on space-time feature depth fusion, which aims to deeply integrate complex space-time dependence of traffic flow to form composite traffic situation awareness and realize multi-period comprehensive traffic flow prediction. Firstly, extracting time sequence characteristics of a multi-time mode through space-time information comprehensive characterization and constructing a real space relationship with an appearance. In order to further construct a digital urban traffic flow prediction method integrating multiple scales and multiple granularity comprehensive elements, the invention provides a comprehensive traffic flow prediction model deeply integrating short-term time-space dependence aiming at short-term traffic flow prediction, improves sensitivity to instantaneous flow change, fully models short-term dynamic change of traffic flow in a multifunctional airspace node, and designs a model integrating deep airspace deconstruction and time sequence characteristics for medium-long-term traffic prediction to deeply extract complex traffic modes. Simulation results show that the method provided by the invention is superior to the existing traffic flow prediction technology.","assignee":"Harbin University of Science and Technology","inventors":["赵中楠","谢旭","王钺"],"publication_date":"2025-10-17","filing_date":"2024-11-12","priority_date":"2024-11-12","cpc_codes":["G","G08","G08G","G08G1/00","G08G1/01","G08G1/0104","G08G1/0125","G08G1/0129","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G08","G08G","G08G1/00","G08G1/01","G08G1/0104","G08G1/0137"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN119541195B/en"},{"publication_number":"AU2025238013A1","title":"Determining a kinematic state of a load handling device in a storage system","abstract":"A load handling device in a storage system is arranged to selectively move in at least one of the X and/or Y directions on tracks and to handle a container, the load handling device comprising a plurality of wheels, and a slip control manager arranged to manage the slip of the load handling device.","assignee":"Ocado Innovation Ltd","inventors":["Thomas BRUEN","Chrysanthos DIMITROPOULOS","Stefano RAIMONDI COMINESI","Ernesto TRIPODI"],"publication_date":"2025-10-16","filing_date":"2025-09-25","priority_date":"2021-08-20","cpc_codes":["B","B65","B65G","B65G1/00","B65G1/02","B65G1/04","B65G1/0464","B","B65","B65G","B65G1/00","B65G1/02","B65G1/04","B65G1/0478","B","B65","B65G","B65G1/00","B65G1/02","B65G1/04","B65G1/0492","B","B65","B65G","B65G1/00","B65G1/02","B65G1/04","B65G1/06","B65G1/065","B","B65","B65G","B65G1/00","B65G1/02","B65G1/04","B65G1/137","B65G1/1373","B65G1/1375","B","B65","B65G","B65G1/00","B65G1/02","B65G1/04","B65G1/137","B65G1/1373","B65G1/1378","G","G05","G05D","G05D1/00","G05D1/20","G05D1/24","G05D1/244","G05D1/2446","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","B","B65","B65G","B65G2201/00","B65G2201/02","B65G2201/0235","B","B65","B65G","B65G2203/00","B65G2203/02","B65G2203/0266","B","B65","B65G","B65G2203/00","B65G2203/04","B65G2203/042","G","G05","G05D","G05D2101/00","G05D2101/22","G","G05","G05D","G05D2105/00","G05D2105/20","G05D2105/28","G","G07","G07C","G07C5/00","G07C5/02"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025238013A1/en"},{"publication_number":"AU2025237978A1","title":"Systems and methods for delivery of digital biomarkers and genomic panels","abstract":"41 Systems and methods are disclosed for receiving one or more digital images associated with a tissue specimen, a related case, a patient, and/or a plurality of clinical information, determining one or more of a prediction, a recommendation, and/or a plurality of data for the one or more digital images using a machine learning system, the machine learning system having been trained using a plurality of training images, to predict a biomarker and a plurality of genomic panel elements, and determining, based on the prediction, the recommendation, and/or the plurality of data, whether to log an output and at least one visualization region as part of a case history within a clinical reporting system.","assignee":"Paige AI Inc","inventors":["Thomas Fuchs","Leo Grady","Christopher Kanan","Jason Locke","Peter SCHUEFFLER","Jilian SUE"],"publication_date":"2025-10-16","filing_date":"2025-09-25","priority_date":"2020-01-28","cpc_codes":["G","G06","G06N","G06N20/00","G","G16","G16H","G16H30/00","G16H30/20","G","G16","G16H","G16H30/00","G16H30/40","G","G16","G16H","G16H50/00","G16H50/20"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025237978A1/en"},{"publication_number":"AU2025238047A1","title":"Quality Assurance Systems Based On Speaker Intent Detection","abstract":"59 A semantic similarity based configurable system for automatic scenario detection in customer- agent conversations is disclosed. The system understands intent from the vector space semantic similarity between speaker sentences, which is agnostic to the use of synonyms and tolerates a large amount of paraphrasing. This approach scales easily to a large number of customers and can be fed more data to increase accuracy and precision. Furthermore, the system is configurable in real-time so that the client is able to control which intents are detected and how. In some embodiments, the semantic similarity based configurable system comprises a scenario detection system, a conversation tag system, a bi-encoder, and a cross-encoder, where the scenario detection system receives inputs of sample phrases and customer-agent utterances and generates results. The sample phrases may be phrases and keywords that describe a scenario expressing the behavior of a customer or call agent.","assignee":"Ujwal Inc","inventors":["Sumeet Khullar","Ashish Nagar","Tanul Singh","Madhur SINGHAL","Abhimanyu TALWAR"],"publication_date":"2025-10-16","filing_date":"2025-09-25","priority_date":"2023-01-31","cpc_codes":["G","G10","G10L","G10L15/00","G10L15/08","G10L15/18","G10L15/1815","G","G10","G10L","G10L15/00","G10L15/08","G10L15/18","G10L15/1822","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N7/00","G06N7/01","G","G10","G10L","G10L15/00","G10L15/08","G10L15/16","G","G06","G06N","G06N3/00","G06N3/12","G06N3/126"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025238047A1/en"},{"publication_number":"AU2025237967A1","title":"System For Interactive Sports Analytics Using Multi-Template Alignment And Discriminative Clustering","abstract":"A system is described for interactively analyzing plays of a sporting event based on real- world positional tracking data. Using positional information regarding the players and/or ball and/or other objects obtained from a tracking system, along with identified event data and contextual information, the system processes a library of plays (e.g., one or more seasons’ worth of a league’s contests) into a searchable database of plays using multiple alignment templates and discriminative clustering techniques. A user interface is described for interacting with the database in a graphical manner, whereby users can query a graphical depiction of a play and receive the most similar plays from the library, along with statistical information relating to the plays. The user interface further permits the user to modify the query graphically (e.g., moving or exchanging players, ball trajectories, etc.) and obtain updated statistical information for comparison.","assignee":"Stats LLC","inventors":["Patrick Lucey","Long SHA","Xinyu WEI"],"publication_date":"2025-10-16","filing_date":"2025-09-24","priority_date":"2015-12-14","cpc_codes":["A","A63","A63F","A63F13/00","A63F13/30","A63F13/35","A","A63","A63F","A63F13/00","A63F13/20","A63F13/21","A63F13/216","A","A63","A63F","A63F13/00","A63F13/25","A","A63","A63F","A63F13/00","A63F13/80","A63F13/812","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N7/00","G06N7/01","A","A63","A63F","A63F13/00","A63F13/60","A63F13/65","G","G06","G06F","G06F16/00","G06F16/70","G06F16/78","G06F16/7867"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025237967A1/en"},{"publication_number":"US20250323822A1","title":"Real-time monitoring ecosystem","abstract":"A network system to provide real-time integration and processing of user data with infrastructure data to generate solutions to user pain points. Real-time user data, including feedback and interactions, is generally not uniform and overwhelmingly large. The system provides solutions to user pain-points at scale, which, in some instances, may be unknown to the service provider. The system does so by contextually linking user data and categorizing it into standardized taxonomies. The infrastructure data is then analyzed against the taxonomies by the system's AI/ML network. The system then provides one or more pain point identifications and solutions. The system may also provide an interface to visualize the taxonomies, pain points, and trend analysis of the pain points.","assignee":"Citibank NA","inventors":["Japan Mehta","Abhishek Seth","Siva Koti Reddy Malapati"],"publication_date":"2025-10-16","filing_date":"2025-06-24","priority_date":"2023-03-31","cpc_codes":["G","G06","G06N","G06N7/00","G06N7/01","H","H04","H04L","H04L41/00","H04L41/06","H04L41/0631","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/01","H","H04","H04L","H04L41/00","H04L41/16"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250323822A1/en"},{"publication_number":"KR20250148695A","title":"Q-Predictor HTC™: Q-Block Based High-Throughput Time-Series Prediction Computing System and Method","abstract":"본 발명은 Q-Block 구조를 이용하여 슈퍼컴퓨터 수준의 시계열 예측을 실시간 또는 준실시간으로 구현하는 연산 시스템 및 방법에 관한 것이다. 구체적으로, 입력된 시계열 데이터는 단위 시간(TSU, 122) 단위로 단위큐브(110)에 매핑되어 적층되며, 각 단위큐브에는 위상(φ) 및 변화량(Δ) 메타데이터가 기록된다. 전이 모듈(150)은 중심 단위큐브(121) 간의 anchor-to-anchor 전이 조건을 산출하고, 예측 모듈(160)은 이를 기반으로 미래 TSU 시점의 상태를 산출한다. 강화학습 모듈(170)은 예측 결과와 실제 데이터를 비교하여 파라미터를 자동 갱신함으로써, 환경 변화나 데이터 분포의 변동에 적응할 수 있도록 한다. 본 발명은 기상, 에너지, 금융 등 대규모 시뮬레이션 환경에 적용 가능하며, 종래 슈퍼컴퓨터 기반 수치예측 대비 연산량을 대폭 절감한다. 또한 GPU 기반 병렬 아키텍처뿐만 아니라 NPU·VPU·FPGA 등 엣지 전용 하드웨어와도 자연스럽게 결합되어, 클라우드·데이터센터뿐 아니라 스마트폰, 차량용 단말, 드론 모듈 등 모바일·엣지 환경에서도 저전력 고효율 예측을 실현함으로써 비용 절감과 응답 속도의 향상을 동시에 제공한다. The present invention relates to a computing system and method for implementing supercomputer-level time series forecasting in real time or near real time using a Q-Block structure. Specifically, input time series data is mapped and stacked on unit cubes (110) in units of time units (TSUs, 122), and phase (φ) and change amount (Δ) metadata are recorded for each unit cube. The transition module (150) calculates anchor-to-anchor transition conditions between central unit cubes (121), and the prediction module (160) calculates a state at a future TSU point in time based on the conditions. The reinforcement learning module (170) automatically updates parameters by comparing the predicted results with actual data, thereby enabling adaptation to environmental changes or fluctuations in data distribution. The present invention can be applied to large-scale simulation environments such as weather, energy, and finance, and significantly reduces the amount of computation compared to conventional supercomputer-based numerical forecasting. In addition, it is naturally combined with edge-specific hardware such as NPU, VPU, and FPGA as well as GPU-based parallel architecture, thereby realizing low-power, high-efficiency prediction not only in cloud and data centers but also in mobile and edge environments such as smartphones, vehicle terminals, and drone modules, thereby simultaneously providing cost reduction and improved response speed.","assignee":"강성운","inventors":["강성운"],"publication_date":"2025-10-14","filing_date":"2025-09-22","priority_date":"2022-02-08","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","H","H10","H10K","H10K59/00","H10K59/10","H10K59/12","H10K59/131","H","H10","H10D","H10D86/00","H10D86/40","H10D86/441","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06Q","G06Q20/00","G06Q20/30","G06Q20/36","G","G06","G06Q","G06Q20/00","G06Q20/30","G06Q20/36","G06Q20/367","G06Q20/3678","G","G06","G06Q","G06Q30/00","G06Q30/06","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0613","G06Q30/0619","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0641","G06Q30/0643","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/16","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/16","G06Q50/163","G","G06","G06T","G06T13/00","G06T13/20","G06T13/40","G","G06","G06T","G06T17/00","G06T17/05","G","G06","G06T","G06T17/00","G06T17/10","G","G06","G06T","G06T19/00","G","G06","G06T","G06T19/00","G06T19/003","G","G06","G06T","G06T19/00","G06T19/20","H","H10","H10D","H10D86/00","H10D86/40","H10D86/451","H","H10","H10D","H10D86/00","H10D86/40"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250148695A/en"},{"publication_number":"CN120780489A","title":"Industrial defect detection task scheduling method, device, equipment and medium","abstract":"The application discloses an industrial defect detection task scheduling method, device, equipment and medium, which relate to the technical field of artificial intelligence and are used for carrying out data acquisition and preprocessing on an edge artificial intelligence task for industrial defect detection, carrying out feature extraction on preprocessed data to obtain a feature vector, screening a target computing core matched with the feature vector, distributing priority to the edge artificial intelligence task, carrying out energy efficiency optimization on the target computing core based on the target edge artificial intelligence task, adding the target edge artificial intelligence task to a task queue based on the priority, acquiring load parameters of the optimized computing core, judging whether the load parameters are larger than a preset threshold, and if so, migrating the task to be scheduled to the computing core to be scheduled to complete the scheduling of the industrial defect detection task, thereby solving the problems of unreasonable scheduling of the industrial defect detection task, low computing resource waste and low energy efficiency of the computing equipment under a mixed multi-core architecture and realizing reasonable allocation of the task on the mixed multi-core architecture.","assignee":"Inspur Enterprise Cloud Technology Shandong Co ltd","inventors":["吕友","毕姗姗","王凤春"],"publication_date":"2025-10-14","filing_date":"2025-09-12","priority_date":"2025-09-12","cpc_codes":["G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5027","G06F9/5038","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120780489A/en"},{"publication_number":"CN120783240A","title":"Multi-temporal remote sensing data-based rice planting area identification method, device, equipment and storage medium","abstract":"The application provides a method, a device, equipment and a storage medium for identifying a rice planting area based on multi-time-phase remote sensing data, wherein the method comprises the steps of obtaining the multi-time-phase remote sensing data according to a rice weather period; determining a spectrum index sequence of multi-time-phase remote sensing data, constructing a spectrum index time sequence curve, generating planting characteristic data in a target area according to the spectrum change curve, determining a spectrum time sequence template matched with a rice growth rule according to the planting characteristic data, carrying out matching analysis on each plot planting mode, marking a plot set of rice planting mode characteristics, carrying out cross comparison on the plot set and a rice planting reference sample to generate a rice planting mode sample set, constructing a rice planting area identification model based on the rice planting mode sample set, analyzing remote sensing data of plots to be identified in a research area according to the rice planting area identification model, and determining a rice planting area identification result. By implementing the scheme of the application, the accuracy of rice identification can be effectively improved.","assignee":"Anhui Feiwei Information Technology Co ltd; Feiwei Shuzhi Technology Wuxi Co ltd; Feiwei Information Technology Co ltd","inventors":["罗顶林","郭美春","岳燕","胡剑锋","曾宁","梁栋","董怀龙","刘慧�","黄洋","兰娟"],"publication_date":"2025-10-14","filing_date":"2025-09-12","priority_date":"2025-09-12","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/10","G06V20/13","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06V","G06V10/00","G06V10/40","G","G06","G06V","G06V10/00","G06V10/40","G06V10/52","G","G06","G06V","G06V10/00","G06V10/40","G06V10/58","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G06V10/765","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/10","G06V20/188","G","G06","G06V","G06V20/00","G06V20/10","G06V20/194"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120783240A/en"},{"publication_number":"KR20250148525A","title":"Youth welfare-based crisis response automation system and method","abstract":"본 발명은 청소년 복지 분야의 위기 대응 자동화 시스템에 관한 것이다. 본 발명의 시스템은 모바일 단말에서 위기 발생 신고를 접수하고, 인공지능 기반 분석으로 위험도를 분류한 후, 쉼터·의료·복지 기관 자원의 가용성을 확인하여 우선순위에 따라 매칭된 기관과 API 연계를 통해 자동으로 연계한다. 또한 법정 프로토콜에 따른 조치를 자동 개시하고, 모든 과정을 감사 로그로 기록하며 예산 정산 보고를 생성한다. 이를 통해 위기 상황에 대한 대응의 속도와 정확성을 높이고, 청소년 복지 서비스 제공의 효율성을 향상시킬 수 있다. The present invention relates to an automated crisis response system in the youth welfare field. The system receives crisis reports from mobile devices, classifies risk levels through AI-based analysis, then checks the availability of resources at shelters, medical facilities, and welfare organizations. It then automatically connects the respondent to the appropriate organization based on priority through API linkage. Furthermore, it automatically initiates actions in accordance with legal protocols, records all processes in an audit log, and generates a budget settlement report. This system can enhance the speed and accuracy of crisis response and improve the efficiency of youth welfare services.","assignee":"조민기","inventors":["조민기"],"publication_date":"2025-10-14","filing_date":"2025-09-12","priority_date":"2025-09-12","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/22","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G06Q10/06311","G06Q10/063112","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0635","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/103","G","G08","G08B","G08B27/00","G08B27/001","H","H04","H04W","H04W4/00","H04W4/02","H04W4/029"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250148525A/en"},{"publication_number":"CN120781950A","title":"Automatic knowledge graph construction system and method based on LANGGRAPH workflow","abstract":"The invention discloses a knowledge graph automatic construction system and method based on LANGGRAPH workflow, which relate to the field of knowledge graphs and comprise a multi-source heterogeneous data acquisition module, a knowledge extraction and graph construction module, a Wikidata structured attribute integration module and an entity fusion and disambiguation processing module, wherein the multi-source heterogeneous data acquisition module is used for acquiring unstructured text data of wikipedia of a target entity and structured attribute data of Wikidata according to a name list, the knowledge extraction and graph construction module is used for carrying out knowledge extraction and graph construction on the unstructured text data of wikipedia based on LANGGRAPH workflow so as to obtain a preliminary graph structure and storing the preliminary graph structure into a graph database, the Wikidata structured attribute integration module is used for obtaining an intermediate graph structure according to the structured attribute data of Wikidata, and the entity fusion and disambiguation processing module is used for carrying out entity fusion and disambiguation processing on the intermediate graph structure in the graph database so as to obtain a final knowledge graph structure. The invention improves the accuracy and the automation degree of knowledge graph construction.","assignee":"Data Space Research Institute","inventors":["张新宇","陈伟健","陈佳佳","邱阳","王玮琦","董文祥","何刘飞"],"publication_date":"2025-10-14","filing_date":"2025-09-12","priority_date":"2025-09-12","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G06F16/9024","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G","G06","G06F","G06F40/00","G06F40/20","G06F40/279","G06F40/289","G06F40/295","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","Y","Y02","Y02D","Y02D10/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120781950A/en"},{"publication_number":"CN120782682A","title":"Endoscope video image enhancement method and system based on fog-free frame screening","abstract":"本发明涉及图像增强技术领域，且公开了一种基于无雾帧筛选的内窥镜视频图像增强方法及系统。该方法及系统通过引入基于深度学习的无雾帧筛选机制，能够从原始内窥镜视频中自动筛选出无雾且清晰的参考帧，避免了人工标注和低效随机采样，结合物理模型与加雾模拟算法，可生成高质量的加雾图像，构建针对内窥镜场景的加雾到无雾配对训练数据集，为后续去雾模型的训练提供了可靠支撑，去雾模型部分根据视频分辨率智能选择最优算法，针对1080p视频采用AOD算法，针对4K视频采用暗通道方法，实现了对不同分辨率视频的自适应处理，确保了处理结果的高效性与一致性，同时支持用户自定义输出帧率和分辨率，通过动态调整缩放策略，满足不同临床应用场景的需求。 The present invention relates to the field of image enhancement technology, and discloses an endoscopic video image enhancement method and system based on fog-free frame screening. The method and system can automatically screen out fog-free and clear reference frames from the original endoscopic video by introducing a fog-free frame screening mechanism based on deep learning, avoiding manual labeling and inefficient random sampling. Combining the physical model with the fogging simulation algorithm, it can generate high-quality fogged images, construct a fogged to fog-free pairing training data set for endoscopic scenes, and provide reliable support for the training of subsequent defogging models. The defogging model part intelligently selects the optimal algorithm according to the video resolution, adopts the AOD algorithm for 1080p video, and adopts the dark channel method for 4K video, realizing adaptive processing of videos with different resolutions, ensuring the efficiency and consistency of the processing results, and supporting users to customize the output frame rate and resolution, and meet the needs of different clinical application scenarios by dynamically adjusting the scaling strategy.","assignee":"Zhoushan hospital","inventors":["董金良","裘文汇","蔡嘉麒","孙健哲","温程远","刘志松","杨堃"],"publication_date":"2025-10-14","filing_date":"2025-09-12","priority_date":"2025-09-12","cpc_codes":["G","G06","G06T","G06T5/00","G06T5/77","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06T","G06T5/00","G06T5/50","G","G06","G06T","G06T5/00","G06T5/60","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120782682A/en"},{"publication_number":"CN114138971B","title":"A Genetic Algorithm-Based Extremely Large Multi-Label Classification Method","abstract":"The invention discloses a genetic algorithm-based maximum multi-label classification method, which comprises the steps of converting texts into word vectors and calculating average word vectors of all sample texts, clustering the sample texts in a text word vector space by using a k-means algorithm, selecting a plurality of neighbor labels in clusters corresponding to all samples, extracting joint characteristic representations of the samples and semantic labels, namely projecting sample neighbor label vectors into a low-dimensional space to obtain low-dimensional characteristic representations of the neighbor labels, combining sample text characteristics extracted by a convolutional neural network to obtain joint characteristic representations of the samples and the semantic labels, designing a loss value of joint characteristic representations of network learning measurement samples and the semantic labels, and creatively guiding a genetic algorithm to find the semantic labels which are best matched with new samples by using the loss value as prediction labels of the samples. According to the method, the real labels of the sample are indirectly restored through the Hamming distance between the regression sample prediction labels and the real labels, so that huge calculation resources and time resource consumption in the problem of classifying the vast majority of labels are avoided.","assignee":"Jiangsu University","inventors":["李丽莎","马忠臣","毛启容","成鑫","陈松灿"],"publication_date":"2025-10-14","filing_date":"2021-11-29","priority_date":"2021-11-29","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/35","G","G06","G06F","G06F18/00","G06F18/20","G06F18/23","G06F18/232","G06F18/2321","G06F18/23213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G06F40/216","G","G06","G06N","G06N3/00","G06N3/12","G06N3/126"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN114138971B/en"},{"publication_number":"US12443881B1","title":"Apparatus and methods to provide a learning agent with improved computational applications in complex real-world environments using machine learning","abstract":"Embodiments disclosed include a method comprising building a machine learning (ML) model that includes a digital agent configured to navigate a digital environment, implementing a navigation of the digital agent over a period of time and following a path within the digital environment, that includes a set of states. Upon reaching each state, the digital agent is configured to acquire the value associated with that state. The method includes determining a signal representing an acquisition of values by the digital agent over the period of time, computing an emotive state of the digital agent and an arousal state of the digital agent based on the signal representing the acquisition of values, and determining a goal for the digital agent to achieve an improved overall performance in acquiring cumulative rewards. The method includes implementing a real version of an action of the digital agent in a real environment.","assignee":"Substrate Artificial Intelligence Sa","inventors":["James Brennan WORTH"],"publication_date":"2025-10-14","filing_date":"2022-09-02","priority_date":"2021-09-02","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N7/00","G06N7/01"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12443881B1/en"},{"publication_number":"CN120781920A","title":"Large model fine tuning method and related device based on causal graph and thinking chain enhancement","abstract":"A big model fine tuning method and a relevant device based on causal graph and thinking chain enhancement relate to the technical field of big model fine tuning in the power industry, and the method comprises the steps of performing causal mining on power equipment data to construct a power equipment causal graph containing causal weight information; the method comprises the steps of disassembling input of a large model into a thinking chain, correspondingly generating chain-type causal pairs according to the thinking chain, carrying out path retrieval matching and causal consistency checking on the chain-type causal pairs and a constructed causal graph of the power equipment to realize alignment of an reasoning process, exciting a reinforcement learning process through an alignment result of the reasoning process, optimizing a reinforcement learning rewarding model which is built in advance, restricting the thinking chain generation process, guiding the large model to generate the thinking chain under a causal constraint condition, and realizing fine adjustment of the large model. The invention embeds causal reasoning and causality into the reinforcement learning feedback process of the fine adjustment of the large model, so that the large model can learn basic causal reasoning rules and can promote the logics, the interpretations and the robustness of thinking chain reasoning.","assignee":"China Electric Power Research Institute Co Ltd CEPRI","inventors":["马震媛","张英强","梁潇","张中浩","唐鹏飞","龙天航","宋博川"],"publication_date":"2025-10-14","filing_date":"2025-06-30","priority_date":"2025-06-30","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06N","G06N7/00","G06N7/01"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120781920A/en"},{"publication_number":"CN120030973B","title":"Integrated circuit layout design teaching auxiliary method","abstract":"The application discloses an integrated circuit layout design teaching auxiliary method, which relates to the field of integrated circuits and comprises the steps of receiving a circuit diagram creation instruction, generating a circuit diagram drawing interface according to the circuit diagram creation instruction, identifying the type and the connection relation of circuit devices in the circuit diagram drawing interface, generating a corresponding circuit schematic diagram data structure by adopting a graph neural network GNN algorithm based on message passing according to the identified circuit device type and the connection relation, converting the circuit schematic diagram data structure into a layout data structure, wherein the layout data structure comprises geometric parameters and layout position data of layout elements, and generating a circuit layout by utilizing a reinforcement learning algorithm based on the graph neural network according to the layout data structure. Aiming at the low design efficiency of the integrated circuit layout in the prior art, the design efficiency is improved by constructing a multi-objective optimized Markov decision process, utilizing deep learning, reinforcement learning algorithm and the like.","assignee":"Qingdao Qingruan Jingzun Microelectronics Technology Co ltd","inventors":["马艳辉","张侠","李克坚","苑芳"],"publication_date":"2025-10-14","filing_date":"2025-02-06","priority_date":"2025-02-06","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/30","G06F30/39","G06F30/392","G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06F","G06F30/00","G06F30/30","G06F30/39","G06F30/398","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N7/00","G06N7/01"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120030973B/en"},{"publication_number":"CN120781945A","title":"Relation perception gating neural network link prediction method oriented to coal rock knowledge graph","abstract":"本发明涉及一种面向煤岩知识图谱的关系感知门控神经网络链路预测方法，属于知识图谱和人工智能技术领域。包括以下步骤：加载煤岩知识图谱数据集；基于知识图谱数据集生成提示图并获取提示图嵌入表示；利用提示图嵌入表示对知识图谱数据集中的关系嵌入表示进行初始化，并基于N层GIN图神经网络结构作为知识图谱消息传递架构更新实体表示；利用更新后的实体表示为候选实体分配分数进行预测推理；通过设计关系感知神经元RPRU模块来控制信息流动，平衡和结合局部与全局特征，获取更加合理的提示示例的嵌入表示；采用N层GIN神经网络结构设计消息传递机制，以更高效地捕捉图结构信息，从而提高图提示信息在训练和推理过程中的有效利用。 The present invention relates to a relationship-aware gated neural network link prediction method for coal and rock knowledge graphs, belonging to the fields of knowledge graphs and artificial intelligence technologies. The method comprises the following steps: loading a coal and rock knowledge graph dataset; generating a prompt graph based on the knowledge graph dataset and obtaining a prompt graph embedding representation; initializing the relationship embedding representation in the knowledge graph dataset using the prompt graph embedding representation, and updating the entity representation using an N-layer GIN graph neural network structure as a knowledge graph message passing architecture; using the updated entity representation to assign scores to candidate entities for predictive reasoning; controlling information flow by designing a relationship-aware neuron (RPRU) module, balancing and combining local and global features, and obtaining a more reasonable embedding representation of prompt examples; and designing a message passing mechanism using the N-layer GIN neural network structure to more efficiently capture graph structure information, thereby improving the effective use of graph prompt information in the training and reasoning processes.","assignee":"Linyi University","inventors":["王星","朱仰瑞","姚双龙","陈吉","刘烨","杨亭","贾俊华","张问银","王海峰","刘志强"],"publication_date":"2025-10-14","filing_date":"2025-09-12","priority_date":"2025-09-12","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N5/00","G06N5/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120781945A/en"},{"publication_number":"US12443850B2","title":"Trainable differential privacy for machine learning","abstract":"Technologies are provided for training machine learning models using a differential privacy mechanism. Training data can be transformed using a differential privacy mechanism that comprises a trainable confidence parameter. The transformed training data can be used to generate class predictions using the machine learning model. A class prediction loss can be determined based on differences between the class predictions and actual classes for the training data. A membership inference loss can also be determined based on predictions that example records in the transformed data set are actual members of the original training data. The membership inference loss and the class prediction loss can be combined to generate a classifier loss that can be used to update the machine learning model and to update the trainable confidence parameter of the differential privacy mechanism. The training can be repeated multiple times until the combined classifier loss falls below a specified threshold.","assignee":"SAP SE","inventors":["Anderson Santana de Oliveira","Caelin Kaplan"],"publication_date":"2025-10-14","filing_date":"2021-07-19","priority_date":"2021-07-19","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06F","G06F21/00","G06F21/60","G06F21/62","G06F21/6218","G06F21/6245","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06N","G06N5/00","G06N5/04"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12443850B2/en"},{"publication_number":"US12443839B2","title":"Hyperparameter transfer via the theory of infinite-width neural networks","abstract":"Systems and method are provided that are directed to tuning a hyperparameter associated with a small neural network model and transferring the hyperparameter to a large neural network model. At least one neural network model may be received along with a request for one or more tuned hyperparameters. Prior to scaling the large neural network, the large neural network is parameterized in accordance with a parameterizing scheme. The large neural network is then scaled and reduced in size such that a hyperparameter tuning process may be performed. A tuned hyperparameter may then be provided to a requestor such that the hyperparameter can be directly input into the large neural network. By tuning a hyper parameter using a small neural network, significant computation cycles and energy may be saved.","assignee":"Microsoft Technology Licensing LLC","inventors":["Jingfeng HU","Ge Yang","Xiaodong Liu","Jianfeng Gao"],"publication_date":"2025-10-14","filing_date":"2020-08-21","priority_date":"2020-08-21","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12443839B2/en"},{"publication_number":"US12445677B2","title":"Small and fast video processing networks via neural architecture search","abstract":"Generally, the present disclosure is directed to a neural architecture search process for finding small and fast video processing networks for understanding of video data. The neural architecture search process can automatically design networks that provide comparable video processing performance at a fraction of the computational and storage cost of larger existing models, thereby conserving computing resources such as memory and processor usage.","assignee":"Google LLC","inventors":["Anthony J. Piergiovanni","Anelia Angelova","Michael Sahngwon Ryoo"],"publication_date":"2025-10-14","filing_date":"2020-09-16","priority_date":"2019-09-18","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/217","G","G06","G06F","G06F18/00","G06F18/20","G06F18/285","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/086","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/40","H","H04","H04N","H04N21/00","H04N21/40","H04N21/43","H04N21/44","H04N21/44008","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12445677B2/en"},{"publication_number":"CN120611100B","title":"Intent-aware recommendation method based on the fusion of knowledge graph and contrastive learning","abstract":"本发明提供了基于知识图谱与对比学习融合的意图感知推荐方法，涉及人工智能技术领域，该方法通过融合知识图谱和对比学习技术来优化节点表示，从而提升推荐系统对长尾项目的推荐精度。同时，在数据增强的基础上，结合用户意图扩充对比学习中的正样本选择，进一步强化模型在用户意图建模方面的能力，最终提高推荐结果的准确性和相关性。旨在解决现有技术中因知识图谱噪声干扰导致的长尾项目表征质量差、用户意图建模不足以及推荐精度受限的问题。 The present invention provides an intent-aware recommendation method based on the fusion of knowledge graph and contrastive learning, which relates to the field of artificial intelligence technology. The method optimizes node representation by fusing knowledge graph and contrastive learning technology, thereby improving the recommendation accuracy of the recommendation system for long-tail items. At the same time, on the basis of data enhancement, the positive sample selection in contrastive learning is expanded in combination with user intent, further strengthening the model's ability in user intent modeling, and ultimately improving the accuracy and relevance of the recommendation results. It aims to solve the problems in the prior art of poor long-tail item representation quality, insufficient user intent modeling, and limited recommendation accuracy caused by knowledge graph noise interference.","assignee":"Huaqiao University","inventors":["孙成柱","黄智翔","何霆","余宇濠","黎建桥"],"publication_date":"2025-10-14","filing_date":"2025-08-12","priority_date":"2025-08-12","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G06F16/9535","G","G06","G06F","G06F18/00","G06F18/20","G06F18/22","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120611100B/en"},{"publication_number":"US12443845B2","title":"System and method for online time-series forecasting using spiking reservoir","abstract":"This disclosure relates generally to time series forecasting, and, more particularly, to a system and method for online time series forecasting using spiking reservoir. Existing systems do not cater for efficient online time-series analysis and forecasting due to their memory and computation power requirements. System and method of the present disclosure convert a time series value F(t) at time ‘t’ to an encoded multivariate spike train and extracts temporal features from the encoded multivariate spike train by the excitatory neurons of a reservoir, predict a time series value Y(t+k) at time ‘t’ by performing a linear combination of extracted temporal features with read-out weights, compute an error for predicted time series value Y(t+k) with input time series value F(t+k), employs a FORCE learning on read-out weights using the error to reduce error in future forecasting. Feeding a feedback value back to the reservoir to optimize memory of the reservoir.","assignee":"Tata Consultancy Services Ltd","inventors":["Arun George","Dighanchal Banerjee","Sounak DEY","Arijit Mukherjee"],"publication_date":"2025-10-14","filing_date":"2022-11-29","priority_date":"2022-01-14","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/049","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12443845B2/en"},{"publication_number":"KR20250147653A","title":"Method and apparatus for building augmented datasets and developing reinforcement learning-based language model agents for tasks in constrained environments","abstract":"실시예들은 닫힌 세계의 과업을 위한 증강 데이터 세트를 구축하고 강화 학습 기반 언어 모델 에이전트를 개발하기 위한 장치 및 방법을 제시한다. 일 실시예에 따른 상기 방법은, 상기 인공지능 에이전트에서 사용되는 API(application programming interface)와 관련된 정보를 획득하고, 상기 API와 관련된 정보는 입력 텍스트 및 프롬프트를 포함하는 API 정보 및 상기 언어 모델에 대한 정보를 포함하고, 상기 입력 텍스트는, 상기 인공지능 에이전트에서 사용되는 복수의 태스크 각각에 대한 복수의 원본 텍스트 데이터와 상기 복수의 원본 텍스트 데이터의 일부 원본 텍스트 데이터에 해당 태스크의 API 호출과 관련된 주석이 삽입된 샘플 보강 데이터를 포함하고, 상기 프롬프트는, 상기 복수의 태스크 각각에 대해, API가 호출되는 동작을 설명한 자연어 문장을 포함하고, 상기 입력 텍스트, 상기 프롬프트 및 상기 언어 모델에 대한 정보를 기반으로 상기 복수의 원본 텍스트 데이터의 나머지 원본 텍스트 데이터에 대한 API 호출 위치를 상기 복수의 태스크 각각에 대해 결정하고, 상기 나머지 원본 텍스트 데이터에 대한 API 호출 위치는 원본 텍스트 데이터를 구성하는 복수의 토큰 중에서 API 호출을 시작할 확률이 제1 임계 값을 초과하는 토큰의 위치로 결정되고, 상기 나머지 원본 텍스트 데이터에 대한 API 호출 위치에 해당 태스크의 API 호출과 관련된 토큰을 삽입함으로써, 추가 보강 데이터를 생성하고, 상기 샘플 보강 데이터 및 상기 추가 보강 데이터를 기반으로 보강 데이터 세트를 구축하고, 상기 보강 데이터 세트를 기반으로 상기 언어 모델에 대한 파인 튜닝을 통해 강화 학습을 수행하는 단계를 포함할 수 있다. The embodiments present devices and methods for building augmented data sets for closed-world tasks and developing reinforcement learning-based language model agents. According to one embodiment, the method obtains information related to an application programming interface (API) used in the artificial intelligence agent, and the information related to the API includes API information including input text and a prompt and information about the language model, wherein the input text includes a plurality of original text data for each of a plurality of tasks used in the artificial intelligence agent and sample augmented data in which an annotation related to an API call of the corresponding task is inserted into some original text data of the plurality of original text data, and the prompt includes a natural language sentence describing an operation in which an API is called for each of the plurality of tasks, and based on the input text, the prompt, and information about the language model, an API call position for the remaining original text data of the plurality of original text data is determined for each of the plurality of tasks, and the API call position for the remaining original text data is determined as a position of a token among a plurality of tokens constituting the original text data, the probability of starting an API call exceeding a first threshold value, and inserts a token related to the API call of the corresponding task at the API call position for the remaining original text data, thereby generating additional augmented data, and constructing an augmented data set based on the sample augmented data and the additional augmented data, and strengthening the language model through fine tuning based on the augmented data set. It may include steps for performing learning.","assignee":"난춘 주식회사","inventors":["정희재"],"publication_date":"2025-10-13","filing_date":"2025-09-24","priority_date":"2024-10-04","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06F","G06F40/00","G06F40/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/10"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250147653A/en"},{"publication_number":"DE202025105655U1","title":"A federated learning system for data protection-compliant data exchange and collaboration","abstract":"Ein föderiertes Lernsystem (100) für den Datenschutz bei der gemeinsamen Nutzung von Daten und der Zusammenarbeit, bestehend aus: einem Modul zur Datenerfassung und lokalen Vorverarbeitung, das so konfiguriert ist, dass es lokale Datensätze an jedem Teilnehmerknoten bereinigt, normalisiert und standardisiert, ohne Rohdaten extern zu übertragen; ein lokales Modelltrainingsmodul, das so konfiguriert ist, dass es ein maschinelles Lernmodell auf dem vorverarbeiteten lokalen Datensatz trainiert; ein sicheres Modellaktualisierungs- und Verschlüsselungsmodul, das so konfiguriert ist, dass es Modellparameter oder Aktualisierungen vor der Übertragung mithilfe von Techniken zum Schutz der Privatsphäre verschlüsselt und sichert; ein föderiertes Aggregations- und Koordinationsmodul, das so konfiguriert ist, dass es verschlüsselte Aktualisierungen von mehreren Teilnehmerknoten zu einem globalen Modell aggregiert; ein Modul zur Überwachung und Einhaltung des Datenschutzes, das so konfiguriert ist, dass es Datenschutzbudgets und Audit-Protokolle durchsetzt und die Einhaltung der Datenschutzbestimmungen sicherstellt; ein Modul zur Leistungsoptimierung und Ressourcenverwaltung, das so konfiguriert ist, dass es die Kommunikation, Berechnung und Ressourcennutzung über alle Knoten hinweg optimiert; und ein Modul zur globalen Modellbereitstellung und Rückmeldung, das so konfiguriert ist, dass es das aggregierte globale Modell an die Teilnehmer zurückverteilt und Leistungsrückmeldungen für iterative Verbesserungen integriert. A federated learning system (100) for data protection in data sharing and collaboration, consisting of: a module for data acquisition and local preprocessing that is configured to clean, normalize and standardize local data sets at each participating node without transferring raw data externally; a local model training module configured to train a machine learning model on the pre-processed local dataset; a secure model update and encryption module configured to encrypt and secure model parameters or updates before transmission using privacy protection techniques; a federated aggregation and coordination module configured to aggregate encrypted updates from multiple participating nodes into a global model; a module for monitoring and ensuring data protection compliance, configured to enforce data protection budgets and audit protocols and to ensure compliance with data protection regulations; a performance optimization and resource management module configured to optimize communication, computation, and resource utilization across all nodes; and a module for global model delivery and feedback, configured to redistribute the aggregated global model to participants and integrate performance feedback for iterative improvements.","assignee":"Individual","inventors":[],"publication_date":"2025-10-13","filing_date":"2025-09-19","priority_date":"2025-09-19","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098"],"country":"DE","kind":"application","source_url":"https://patents.google.com/patent/DE202025105655U1/en"},{"publication_number":"KR20250144973A","title":"Method and apparatus for generating digital human video based on large model, intelligent agent, electronic device, storage medium, and computer program","abstract":"본 개시는 대규모 모델에 기반한 디지털 휴먼 비디오 생성 방법, 장치, 에이전트, 전자 기기 및 저장 매체를 제공하며, 인공지능 기술 분야에 관한 것으로서, 특히 딥러닝, 대규모 모델, 컴퓨터 비전 등 기술 분야에 관한 것이며, 비디오 생방송, 광고 제작, 전자상거래 판매 등 장면에 적용될 수 있다. 대규모 모델에 기반한 디지털 휴먼 비디오 생성 방법은, 요구 정보를 획득하되, 요구 정보가 지정된 동작 비디오 세그먼트를 설명하기 위한 동작 설명 정보를 포함하고, 동작 비디오 세그먼트가 목표 대상의 지정 동작을 나타내며; 언어 대규모 모델을 이용하여 요구 정보를 처리하여, 목표 대본을 얻되, 목표 대본이 동작 설명 정보와 매칭하는 목표 구술 방송 세그먼트 텍스트를 포함하며; 비전 대규모 모델을 이용하여 목표 대본 및 동작 비디오 세그먼트를 처리하여, 목표 대상이 지정 동작을 수행하는 과정에서, 목표 구술 방송 세그먼트 텍스트에 기초하여 구술 방송을 수행하는 것을 표시하기 위한 목표 비디오를 얻는 것을, 포함한다. The present disclosure provides a method, device, agent, electronic device, and storage medium for generating a digital human video based on a large-scale model, and relates to the field of artificial intelligence technology, and particularly to the fields of deep learning, large-scale models, and computer vision, and can be applied to scenes such as video live broadcasting, advertising production, and e-commerce sales. The method for generating a digital human video based on a large-scale model includes: obtaining request information, wherein the request information includes motion description information for describing a motion video segment for which the request information is specified, and the motion video segment represents a specified motion of a target object; processing the request information using a language large-scale model to obtain a target script, wherein the target script includes a target spoken broadcast segment text that matches the motion description information; and processing the target script and the motion video segment using a vision large-scale model to obtain a target video for representing the target object performing a spoken broadcast based on the target spoken broadcast segment text in the process of performing the specified motion.","assignee":"베이징 바이두 넷컴 사이언스 테크놀로지 컴퍼니 리미티드","inventors":["티엔 우","하이펑 왕","하오 티엔","웬취엔 우","다이 다이","시메이 리우","리 왕","항 저우","총 가오","췬이 시에","칭창 하오"],"publication_date":"2025-10-13","filing_date":"2025-09-17","priority_date":"2025-04-25","cpc_codes":["H","H04","H04N","H04N21/00","H04N21/80","H04N21/85","H04N21/854","H04N21/8545","H","H04","H04N","H04N21/00","H04N21/80","H04N21/81","H04N21/8126","G","G06","G06T","G06T13/00","G06T13/20","G06T13/40","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06T","G06T13/00","G06T13/20","G06T13/205","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/40","G06V20/48","G","G06","G06V","G06V20/00","G06V20/40","G06V20/49","G","G06","G06V","G06V40/00","G06V40/20","G","G10","G10L","G10L13/00","G10L13/08","G","G10","G10L","G10L13/00","G10L13/08","G10L13/10","G","G11","G11B","G11B27/00","G11B27/02","G11B27/031","G","G11","G11B","G11B27/00","G11B27/10","H","H04","H04N","H04N21/00","H04N21/20","H04N21/23","H04N21/234","H04N21/2343","H04N21/234336","H","H04","H04N","H04N21/00","H04N21/40","H04N21/43","H04N21/44","H04N21/4402","H04N21/440236","H"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250144973A/en"},{"publication_number":"KR20250144343A","title":"Q-Guardian Optimizer™ : System and Method for Q-Block Security Policy Optimization Using Reinforcement Learning Reward Function","abstract":"본 발명은 Q-Block 기반 보안·복원 시스템에 강화학습 기반 정책 최적화 모듈을 결합하여, 무결성 검증 및 복원 성능을 자율적으로 향상시키는 기술에 관한 것이다. Intra-Block(200)과 Inter-Block(300) 검증을 통해 산출된 성능 지표(FAR, FRR, Cost, Delay)는 '상태(State)'로 정의되어 정책 최적화 모듈(600)에 입력된다. 본 모듈은 강화학습 보상함수를 통해 파라미터 {w1~w4}를 동적으로 조정하고, 그 결과를 자율 복원 모듈(400)의 가중치 분배에 실시간 반영한다. 복원 결과와 성능 지표는 다시 보상값 R로 계산되어 피드백 루프를 형성하며, 시스템은 공격 패턴 변화에도 지속적으로 적응·학습할 수 있다. 이 구조는 보안·복원 순환 구조(700)의 핵심 요소로서, 자율주행, UAM, 의료 데이터 보안, 금융 거래 보호 등 다양한 산업 분야에 적용 가능하다. The present invention relates to a technology for autonomously improving integrity verification and restoration performance by combining a reinforcement learning-based policy optimization module with a Q-Block-based security and restoration system. Performance indicators (FAR, FRR, Cost, Delay) calculated through Intra-Block (200) and Inter-Block (300) verification are defined as \"States\" and input to a policy optimization module (600). This module dynamically adjusts parameters {w1~w4} through a reinforcement learning reward function and reflects the results in real time in the weight distribution of the autonomous restoration module (400). The restoration results and performance indicators are then calculated as a reward value R, forming a feedback loop, so that the system can continuously adapt and learn even when attack patterns change. This structure is a core element of a security and restoration cyclical structure (700) and can be applied to various industrial fields such as autonomous driving, UAM, medical data security, and financial transaction protection.","assignee":"강성운","inventors":["강성운"],"publication_date":"2025-10-10","filing_date":"2025-09-17","priority_date":"2022-02-08","cpc_codes":["H","H04","H04L","H04L63/00","H04L63/20","H","H10","H10K","H10K59/00","H10K59/10","H10K59/12","H10K59/131","H","H10","H10D","H10D86/00","H10D86/40","H10D86/441","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06Q","G06Q20/00","G06Q20/30","G06Q20/36","G","G06","G06Q","G06Q20/00","G06Q20/30","G06Q20/36","G06Q20/367","G06Q20/3678","G","G06","G06Q","G06Q30/00","G06Q30/06","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0613","G06Q30/0619","G","G06","G06Q","G06Q30/00","G06Q30/06","G06Q30/0601","G06Q30/0641","G06Q30/0643","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/16","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/16","G06Q50/163","G","G06","G06T","G06T13/00","G06T13/20","G06T13/40","G","G06","G06T","G06T17/00","G06T17/05","G","G06","G06T","G06T17/00","G06T17/10","G","G06","G06T","G06T19/00","G","G06","G06T","G06T19/00","G06T19/003","G","G06","G06T","G06T19/00","G06T19/20","H","H04","H04L","H04L63/00","H04L63/12","H04L63/123","H","H04","H04L","H04L9/00","H04L9/50","H","H10","H10D","H10D86/00","H10D86/40","H10D86/451","H","H10","H10D"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250144343A/en"},{"publication_number":"CN120764987A","title":"An AI-based intelligent supply chain management system for nutritious dry goods","abstract":"The invention discloses an AI-based nourishing dry goods intelligent supply chain management system, which belongs to the technical field of data processing and supply chain management and comprises a data acquisition unit, an AI prediction unit, a game balancing unit and a decision execution unit, wherein the data acquisition unit is used for acquiring historical sales data, weather forecast data, real-time road conditions, cold chain sensor data and market information, the AI prediction unit is used for predicting demand and logistics paths based on the data acquired by the data acquisition unit so as to generate a suggested adjustment amount, the game balancing unit is used for performing multi-main game balancing analysis based on the data acquired by the data acquisition unit and a preset utility function so as to calculate an elastic buffer threshold, and the decision execution unit is used for comparing the suggested adjustment amount generated by the AI prediction unit with the elastic buffer threshold calculated by the game balancing unit.","assignee":"Fangjiapuzi Putian Green Food Co ltd","inventors":["方敏","曹连黄","江南","许荣斌","刘志强","许政睿"],"publication_date":"2025-10-10","filing_date":"2025-09-11","priority_date":"2025-09-11","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0631","G","G06","G06N","G06N5/00","G06N5/04","G06N5/042","G","G06","G06Q","G06Q30/00","G06Q30/02","G06Q30/0201","G06Q30/0202"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120764987A/en"},{"publication_number":"CN120765660A","title":"Egg oiling quality detection method based on machine vision","abstract":"本发明属于图像分析的技术领域，具体涉及一种基于机器视觉的鸡蛋涂油质量检测方法，以解决现有技术中传统机器视觉面临低对比度、高光干扰及信息维度单一的技术问题，包括以下步骤：S1，获取待检测鸡蛋的多光谱偏振图像；S2，得到鸡蛋的轮廓掩码和高光区域掩码；S3，得到有效分析区域；S4，生成表征油膜均匀性的线偏振度特征图；S5，生成表征涂油厚度的光谱角特征图；S6，构建以各超像素为节点、邻接关系为边的区域图；S7，融合超像素节点的初始特征向量；S8，识别并输出判定为涂油缺陷的超像素节点；S9，综合评定鸡蛋的涂油质量等级。实现了对鸡蛋涂油质量的自动化、高精度综合分级。 The present invention belongs to the technical field of image analysis and specifically relates to a method for detecting egg oil coating quality based on machine vision. This method addresses the technical issues of low contrast, high light interference, and a single information dimension faced by conventional machine vision in the prior art. The method comprises the following steps: S1, acquiring a multispectral polarization image of the egg to be inspected; S2, obtaining a contour mask and a highlight area mask of the egg; S3, obtaining an effective analysis area; S4, generating a linear polarization degree feature map representing oil film uniformity; S5, generating a spectral angle feature map representing oil coating thickness; S6, constructing a region map with each superpixel as a node and adjacency as an edge; S7, fusing the initial feature vectors of the superpixel nodes; S8, identifying and outputting superpixel nodes determined to be defective in oil coating; and S9, comprehensively assessing the egg oil coating quality. This method achieves automated, high-precision comprehensive grading of egg oil coating quality.","assignee":"Egg No1 Food Co ltd","inventors":["谢怀斌","刘建华","蔡路路","赵风雷","王瑜"],"publication_date":"2025-10-10","filing_date":"2025-09-11","priority_date":"2025-09-11","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/0002","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T7/00","G06T7/10","G","G06","G06T","G06T7/00","G06T7/60","G","G06","G06V","G06V10/00","G06V10/10","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/60","G06V20/68","G","G01","G01N","G01N21/00","G01N21/84","G01N21/88","G01N21/8851","G01N2021/8887","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10048","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120765660A/en"},{"publication_number":"CN120764412A","title":"A method for predicting foundation pit excavation deformation based on steel sheet pile construction parameters","abstract":"The application provides a foundation pit excavation deformation prediction method based on steel sheet pile construction parameters, which is characterized in that real data are obtained based on construction samples, a neural network model is optimized through the real data, then in-situ data of the steel sheet pile construction process are obtained through inversion of the neural network model, and actual construction is guided through the in-situ data. The application realizes popularization and application of small sample data by combining the construction sample and the neural network model, and gives out data required by construction through design parameters of a construction site.","assignee":"Jinan Yellow River Bridge Constrution Group Co Ltd","inventors":["宋彩霞","宋述生","张雷","王乃镇","苗有雨"],"publication_date":"2025-10-10","filing_date":"2025-09-11","priority_date":"2025-09-11","cpc_codes":["G","G06","G06F","G06F30/00","G06F30/20","G06F30/27","G","G06","G06F","G06F30/00","G06F30/10","G06F30/13","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06T","G06T17/00","G06T17/05","G","G06","G06T","G06T17/00","G06T17/10","G","G06","G06F","G06F2111/00","G06F2111/04","G","G06","G06F","G06F2119/00","G06F2119/14","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120764412A/en"},{"publication_number":"CN120763321A","title":"Multi-agent information processing method, device, electronic device, storage medium and program product based on large model","abstract":"本申请实施例提供了一种基于大模型的多智能体信息处理方法、装置、电子设备、存储介质及程序产品。方法包括：根据询问信息生成意图，提高了对询问信息理解的准确率；根据意图生成提示信息，该提示信息用于指示满足该意图的待执行任务，执行该待执行任务可以准确生成回复信息；针对该待执行任务，使用提示信息调用预设流程生成模型生成信息处理流程，该信息处理流程包括多个处理节点以及多个处理节点的调用顺序，符合待执行任务的处理程序，处理节点用于执行待执行任务，提高了待执行任务的处理准确率；采用工具调用智能体，按照信息处理流程顺次调用工具执行对应处理节点的信息处理，以生成所述询问信息的回复信息，提升了用户的体验感。 The present invention provides a multi-agent information processing method, apparatus, electronic device, storage medium, and program product based on a large model. The method includes: generating an intent based on query information, thereby improving the accuracy of understanding the query information; generating prompt information based on the intent, wherein the prompt information is used to indicate a pending task that satisfies the intent, and executing the pending task can accurately generate a reply information; using the prompt information to call a preset process generation model to generate an information processing process for the pending task, wherein the information processing process includes multiple processing nodes and a calling order of the multiple processing nodes that conforms to the processing procedure of the pending task, wherein the processing nodes are used to execute the pending task, thereby improving the processing accuracy of the pending task; using a tool to call an agent, and sequentially calling the tool according to the information processing process to execute information processing of the corresponding processing nodes to generate a reply information for the query information, thereby improving the user experience.","assignee":"Koubei Shanghai Information Technology Co Ltd","inventors":["杨正壮"],"publication_date":"2025-10-10","filing_date":"2025-09-11","priority_date":"2025-09-11","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3344","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F40/00","G06F40/30","G06F40/35","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120763321A/en"},{"publication_number":"CN120763678A","title":"Customer twin construction method based on multi-source data fusion and large model driving","abstract":"The application provides a client twin construction method based on multi-source data fusion and large model driving, and belongs to the technical field of client relationship management. The method comprises the steps of collecting structured data, unstructured data and semi-structured data, generating a global fusion model, extracting a client basic attribute label, constructing a client knowledge graph based on an entity-relation-entity triple structure, storing client-product and client-service interaction relations, inputting client behavior sequence data into a time sequence fusion converter model, capturing long-term dependence and short-term fluctuation characteristics by using a self-attention mechanism in the time sequence fusion converter model, identifying a behavior mode, outputting a dynamic feature vector containing time evolution information, and carrying out weighted fusion on the client basic attribute label and the dynamic feature vector to construct a client digital twin body. The completeness, timeliness and prediction accuracy of the customer portrait are improved.","assignee":"Inspur General Software Co Ltd","inventors":["刘晓倩","龚全玉","闫弋峰"],"publication_date":"2025-10-10","filing_date":"2025-09-11","priority_date":"2025-09-11","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/24765","G","G06","G06F","G06F16/00","G06F16/90","G06F16/901","G06F16/9024","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/23","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120763678A/en"},{"publication_number":"CN120763555A","title":"A method for analyzing accelerometer vibration rectification errors","abstract":"The invention relates to the technical field of data processing, in particular to an accelerometer vibration rectification error analysis method, which comprises the steps of receiving original current signals of a multipath accelerometer, processing collected data to output a characteristic point coordinate sequence with confidence, combining and calculating a local statistical data set according to an optimization algorithm in a confidence and frequency band distribution characteristic dynamic scheduling knowledge base, inputting the data set into a mechanical dynamics model to execute track simulation, simultaneously inputting a control model reconstruction signal to generate an error correction coefficient matrix by comparing a double-path positioning drift difference, reconstructing a vibration rectification error quantized value by adopting the matrix to output a knowledge base updating instruction and a parameter resetting instruction, importing the quantized value to calculate a theoretical angle deviation, verifying convergence, and triggering neural network retraining by a deviation exceeding threshold. The signal distortion problem caused by smoothing operation in the prior art is solved through a non-smoothing characteristic extraction algorithm and a closed loop feedback mechanism.","assignee":"Beijing Xingjian Changkong Measurement Control Technology Co ltd","inventors":["周淼生","李艳军"],"publication_date":"2025-10-10","filing_date":"2025-09-11","priority_date":"2025-09-11","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G","G01","G01P","G01P21/00","G","G06","G06F","G06F18/00","G06F18/10","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2131","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2433","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","Y","Y02","Y02T","Y02T90/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120763555A/en"},{"publication_number":"CN120768789A","title":"An intelligent network test terminal based on artificial intelligence","abstract":"本发明涉及智能网联测试技术领域，公开了一种基于人工智能的智能网联测试终端，其中一种基于人工智能的智能网联测试方法包括：构建物理约束的场景原型表征空间，将场景抽象为满足物理约束的语义向量；设计物理先验知识编码机制，将物理规则嵌入到场景表征空间；开发基于语义空间插值的新场景生成算法，在原型空间中进行有约束的插值；实现多模态一致性验证模块，确保生成场景在不同传感器模态下的一致性；构建场景可执行性评估器，验证生成场景的物理可行性；本发明提高了测试场景的质量和真实性，显著提升了智能网联系统测试的效率和可靠性，为新型智能网联系统的功能验证和测试提供了有力支持。 The present invention relates to the field of intelligent network connection testing technology, and discloses an intelligent network connection testing terminal based on artificial intelligence, wherein an intelligent network connection testing method based on artificial intelligence includes: constructing a scene prototype representation space of physical constraints, abstracting the scene into a semantic vector that satisfies the physical constraints; designing a physical prior knowledge encoding mechanism, and embedding physical rules into the scene representation space; developing a new scene generation algorithm based on semantic space interpolation, and performing constrained interpolation in the prototype space; implementing a multimodal consistency verification module to ensure the consistency of the generated scene under different sensor modes; constructing a scene executable evaluator to verify the physical feasibility of the generated scene; the present invention improves the quality and authenticity of the test scene, significantly improves the efficiency and reliability of intelligent network connection system testing, and provides strong support for the functional verification and testing of new intelligent network connection systems.","assignee":"Changchun Fenghuolun Technology Co ltd","inventors":["李旭","张诗禹"],"publication_date":"2025-10-10","filing_date":"2025-09-11","priority_date":"2025-09-11","cpc_codes":["H","H04","H04L","H04L43/00","H04L43/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06N","G06N5/00","G06N5/04","G06N5/045","H","H04","H04L","H04L41/00","H04L41/16","H","H04","H04L","H04L43/00","H04L43/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120768789A/en"},{"publication_number":"CN120765071A","title":"A farmhouse occupancy rate prediction method integrating multidimensional data","abstract":"本发明涉及数据预测与智能优化技术，提出一种融合多维数据的农家乐入住率预测方法，采集历史入住率、天气、节假日及周边景点访问量，构建训练集并训练初始时序模型；训练中引入改进的语言教育优化算法，自适应调整卷积核规模与TCN层数，得到第二模型；该算法以学习个体相对最差状态的反向方向推导认知目标位置，构建反馈校正因子以动态调节对引导位置的接受强度，并融合种群分布、认知偏移历史片段与认知可信度交互，形成具备非线性跳跃与随机扰动特性的认知演化轨迹；预测阶段引入误差反馈，依据历史残差方差动态调整输出置信度，并与滑动均值进行置信融合，最终给出入住率预测结果。实际证明本方法可提升针对农家乐入住率预测精度与稳定性。 This invention involves data prediction and intelligent optimization technology. It proposes a method for predicting farmhouse occupancy rates by integrating multidimensional data. The method collects historical occupancy rates, weather, holidays, and visitor numbers to surrounding attractions to construct a training set and train an initial time series model. During training, an improved language education optimization algorithm is introduced to adaptively adjust the convolution kernel size and the number of TCN layers to obtain a secondary model. The algorithm derives the cognitive target position in the reverse direction of the learning individual's relative worst state, constructs a feedback correction factor to dynamically adjust the acceptance intensity of the guided position, and integrates population distribution, historical segments of cognitive deviation, and cognitive credibility to form a cognitive evolution trajectory with nonlinear jumps and random perturbations. Error feedback is introduced in the prediction phase to dynamically adjust the output confidence based on the historical residual variance. This confidence is then integrated with the sliding mean to ultimately produce an occupancy prediction result. This method has been shown to improve the accuracy and stability of farmhouse occupancy rate prediction.","assignee":"Ludong University","inventors":["王鹤涛","王建","王超"],"publication_date":"2025-10-10","filing_date":"2025-09-11","priority_date":"2025-09-11","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G06Q10/06375","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/251","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/12"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120765071A/en"},{"publication_number":"CN120759573A","title":"A method and system for controlling drilling parameters of muddy clay pile foundation","abstract":"The invention discloses a method and a system for regulating drilling parameters of a muddy clay pile foundation, which belong to the technical field of pile foundation drilling construction, wherein the method comprises the steps of collecting and preprocessing historical construction data of drilling of the muddy clay pile foundation, and constructing a data set; the method comprises the steps of constructing and training a machine learning model based on a data set to obtain a drilling parameter time sequence prediction model, collecting actual drilling parameters of a drilling process in real time, comparing the actual drilling parameters with predicted drilling parameters output by the model, determining drilling parameter errors, and dynamically adjusting the drilling parameters of a drilling machine based on the drilling parameter errors. The method of the invention realizes the intelligent regulation and control of key parameters such as drilling pressure, rotating speed, slurry concentration and the like, so as to improve the efficiency and pore-forming quality of pile foundation construction in complex soft soil stratum, reduce construction risk and ensure engineering safety.","assignee":"Southwest University of Science and Technology; China Railway Beijing Engineering Group Co Ltd; Urban Rail Transit Engineering Co Ltd of China Railway Beijing Engineering Group Co Ltd","inventors":["范国铮","刘贵香","邓涛","唐磊","候建林","刘龑","牟廷江","孔超"],"publication_date":"2025-10-10","filing_date":"2025-09-11","priority_date":"2025-09-11","cpc_codes":["E","E21","E21B","E21B44/00","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120759573A/en"},{"publication_number":"CN114139066B","title":"A collaborative filtering recommendation system based on graph neural network","abstract":"本发明公开一种基于图神经网络的协同过滤推荐系统，包括用户‑项目二部图生成模块、嵌入信息生成模块、隐式关系构建模块、数据融合模块和推荐模块；本发明使用基于注意力的记忆网络学习分别构建的用户‑用户和项目‑项目图，以获取相邻对之间的关系信息。该模型同时学习所有三种图形，通过信息融合层统一多属性和隐式关系信息，实现端到端推荐。 This paper discloses a collaborative filtering recommendation system based on graph neural networks. The system comprises a user-item bipartite graph generation module, an embedding information generation module, an implicit relationship construction module, a data fusion module, and a recommendation module. The system uses an attention-based memory network to learn the constructed user-user and item-item graphs to obtain relationship information between adjacent pairs. The model simultaneously learns all three graphs and unifies multi-attribute and implicit relationship information through an information fusion layer, enabling end-to-end recommendations.","assignee":"Chongqing University","inventors":["张瀚文","周魏","文俊浩","杨正益","曾骏","覃梦秋","柳玲","蔡海尼","刘林","廖捷"],"publication_date":"2025-10-10","filing_date":"2021-09-10","priority_date":"2021-09-10","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G06F16/9536","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G06F16/9535","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN114139066B/en"},{"publication_number":"CN120278405B","title":"Crop planting intelligent decision-making reasoning system and method based on multi-source data fusion","abstract":"本发明提供一种基于多源数据融合的作物种植智能决策推理系统及方法，涉及农业种植技术领域，系统包括：多源数据融合模块，用于针对多源数据抽取目标作物生长的关键特征；智能感知模块，用于根据作物种植的多源异构数据构建农业知识图谱；推理决策模块，用于根据农业知识图谱、预设的推理模型以及决策模型对多源数据与关键特征进行决策推理。通过本发明，解决传统农作物种植决策系统决策能力不足，导致种植管理过程中的精准度不高，且决策系统大多在单一领域或特定作物上进行优化，缺乏跨领域的多源数据协同处理机制，导致系统的泛化能力不足的缺陷。 The present invention provides a crop planting intelligent decision-making and reasoning system and method based on multi-source data fusion, relating to the field of agricultural planting technology. The system includes: a multi-source data fusion module for extracting key characteristics of target crop growth from multi-source data; an intelligent perception module for constructing an agricultural knowledge graph based on multi-source heterogeneous crop planting data; and a reasoning and decision-making module for performing decision-making and reasoning on the multi-source data and key characteristics based on the agricultural knowledge graph, a preset reasoning model, and a decision-making model. This invention addresses the shortcomings of traditional crop planting decision-making systems, which have limited decision-making capabilities and result in low accuracy in the planting management process. Furthermore, most decision-making systems are optimized for a single field or specific crop, lacking a cross-domain multi-source data collaborative processing mechanism, resulting in insufficient system generalization capabilities.","assignee":"Research Center of Information Technology of Beijing Academy of Agriculture and Forestry Sciences","inventors":["吴华瑞","朱华吉","顾静秋","刘畅","王菲菲"],"publication_date":"2025-10-10","filing_date":"2025-06-10","priority_date":"2025-06-10","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06Q","G06Q50/00","G06Q50/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120278405B/en"},{"publication_number":"CN120764853A","title":"An intelligent control method for plain river network sluice groups based on deep reinforcement learning","abstract":"本发明公开了一种基于深度强化学习的平原河网闸群智能调控方法，包括搭建流量预测模块、结合领域知识与经验库的知识约束强化学习模块以及多智能体强化学习模块；将率定好的知识约束强化学习模块与流量预测模块、多智能体强化学习模块集成，形成完整的平原河网智能调控模型；基于率定好的平原河网智能调控模型，接入未来气象水文预测数据，快速输出合理的闸群调度方案。优点是：智能生成调度方案簇，高效应对复杂、动态变化的平原河网闸群调度问题，提升闸群调度系统的灵活性、时效性和智能化水平，解决机器学习方法在实际应用中可能存在的“非现实解”问题，增强结果的可解释性和方案的实际可操作性，形成“模型—环境—反馈—调控修正”的自适应闭环。 The present invention discloses a method for intelligent control of gate groups in plain river networks based on deep reinforcement learning, including building a flow prediction module, a knowledge constraint reinforcement learning module that combines domain knowledge and an experience library, and a multi-agent reinforcement learning module; integrating the calibrated knowledge constraint reinforcement learning module with the flow prediction module and the multi-agent reinforcement learning module to form a complete plain river network intelligent control model; based on the calibrated plain river network intelligent control model, accessing future meteorological and hydrological forecast data, and quickly outputting a reasonable gate group scheduling plan. The advantages are: intelligent generation of scheduling plan clusters, efficient response to complex and dynamically changing plain river network gate group scheduling problems, improving the flexibility, timeliness and intelligence level of the gate group scheduling system, solving the \"non-realistic solution\" problem that may exist in the actual application of machine learning methods, enhancing the interpretability of the results and the actual operability of the plan, and forming an adaptive closed loop of \"model-environment-feedback-control correction\".","assignee":"Hohai University HHU","inventors":["师鹏飞","杨昊","肖家清","吕凯","卜王超","刘明轩","周志强"],"publication_date":"2025-10-10","filing_date":"2025-07-14","priority_date":"2025-07-14","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q50/00","G06Q50/06"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120764853A/en"},{"publication_number":"CN120764804A","title":"A content analysis and intervention method for value orientation in ideological and political education","abstract":"本发明提供一种思政教育中价值观导向的内容分析与干预方法，属于教育教学技术领域，本发明首先构建多模态价值观知识图谱作为语义基础，采用预训练的多模态价值观理解模型结合递归神经网络提取显性价值观特征并捕获上下文依赖，通过注意力机制和对比学习识别隐性价值观表达，基于特征向量计算与核心价值观的相似度生成倾向性指数，利用强化学习训练内容干预代理生成教育干预策略，最后采用反事实推理评估不同干预策略效果并优化干预方案，形成了从内容分析到干预实施再到效果评估的完整技术闭环，解决了对思政教育内容中隐性价值观难以量化的方式进行精准识别并实施有效干预的技术问题。 The present invention provides a value-oriented content analysis and intervention method in ideological and political education, belonging to the field of education and teaching technology. The present invention first constructs a multimodal value knowledge graph as a semantic basis, adopts a pre-trained multimodal value understanding model combined with a recursive neural network to extract explicit value features and capture context dependencies, identifies implicit value expressions through an attention mechanism and contrastive learning, generates a tendency index based on the similarity between feature vector calculation and core values, utilizes reinforcement learning to train a content intervention agent to generate an educational intervention strategy, and finally adopts counterfactual reasoning to evaluate the effects of different intervention strategies and optimize the intervention plan, forming a complete technical closed loop from content analysis to intervention implementation to effect evaluation, solving the technical problem of accurately identifying and effectively intervening in the implicit values in ideological and political education content that are difficult to quantify.","assignee":"Qingdao Huanghai University","inventors":["陈丽华","徐泉","沈盼盼","夏从亚","徐彬","谈笑","刘小戈","刘馨","芙柔","李安增","张立军"],"publication_date":"2025-10-10","filing_date":"2025-06-24","priority_date":"2025-06-24","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/04","G06Q10/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0637","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/067","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/20","G06Q50/205"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120764804A/en"},{"publication_number":"CN114548423B","title":"Machine learning attention model featuring omnidirectional processing","abstract":"提供了以全向处理为特征的机器学习注意力模型，其示例实施方式可以被称为来自变换器的全向表示(OMNINET)。在本公开中描述的示例模型中，允许每个令牌关注整个网络中的一些或所有其他令牌中的所有令牌，而不是维持严格水平的接受域。 A machine learning attention model featuring omnidirectional processing is provided, an example implementation of which may be referred to as Omnidirectional Representations from Transformers (OMNINET). In the example model described in this disclosure, each token is allowed to attend to all tokens in some or all other tokens in the entire network, rather than maintaining a strict level of receptive field.","assignee":"Google LLC","inventors":["Y.泰伊","D-C.隽","D.巴赫里","D.A.小梅兹勒","J.P.古普塔","M.德哈尼","P.法姆","V.K.阿里班迪","Z.秦"],"publication_date":"2025-10-10","filing_date":"2022-02-07","priority_date":"2021-02-04","cpc_codes":["G","G06","G06N","G06N20/00","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G06","G06F","G06F40/00","G06F40/40","G06F40/58","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN114548423B/en"},{"publication_number":"CN120763503A","title":"Dual-pathway heterogeneous temporal causal analysis method based on representation embedding and time-delay pairing","abstract":"本发明涉及数据分析与人工智能技术领域，具体涉及一种基于表征嵌入与时滞配对的双通路异构时序因果分析方法，包括：采集动力电池状态数据，得到多尺度特征；基于所述多尺度特征进行迭代优化，使失函数最小化，得到动力电池状态因果图的第一边类型置信概率；确定目标‑解释变量对，从所述多尺度特征中多次随机抽取数据形成采样集；基于采样集对所述目标‑解释变量对进行条件独立性检验，获取动力电池状态因果图的第二边类型置信概率；基于动力电池状态因果图的第一和第二边类型置信概率获取融合因果图并进行无环化处理，得到无环动力电池状态因果图作为分析结果；本发明能够增强因果发现的稳健性、可解释性以及效率。 The present invention relates to the field of data analysis and artificial intelligence technology, and in particular to a dual-path heterogeneous temporal causal analysis method based on representation embedding and time lag pairing, comprising: collecting power battery status data to obtain multi-scale features; performing iterative optimization based on the multi-scale features to minimize the loss function and obtain a first edge type confidence probability of a power battery status causal graph; determining a target-explanatory variable pair and randomly extracting data from the multi-scale features multiple times to form a sampling set; performing a conditional independence test on the target-explanatory variable pair based on the sampling set to obtain a second edge type confidence probability of the power battery status causal graph; obtaining a fused causal graph based on the first and second edge type confidence probabilities of the power battery status causal graph and performing acyclic processing to obtain an acyclic power battery status causal graph as an analysis result. The present invention can enhance the robustness, interpretability and efficiency of causal discovery.","assignee":"Beihang University","inventors":["杨顺昆","吴梦丹","李道颐"],"publication_date":"2025-10-10","filing_date":"2025-06-10","priority_date":"2025-06-10","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/29","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G06F18/24155","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N5/00","G06N5/04","G06N5/045","G","G06","G06N","G06N7/00","G06N7/01"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120763503A/en"},{"publication_number":"CN118940838B","title":"Intent recognition method based on large language model and customer portrait classification model","abstract":"一种基于大语言模型和客户画像分类模型的意图识别方法，方法包括：根据客户ID信息获取客户当前的相关订单和背景信息；客户画像分类模型根据相关订单和背景信息获得第一评分向量；第一评分向量用于指示当前客户画像分类得分；微调大模型对输入的客户对话的上下文对话进行语义识别获得第二评分向量；第二评分向量用于指示客户当前情感反馈得分；微调大模型是大语言模型微调得到的；将第一评分向量和第二评分向量融合得到综合评分向量；综合评分向量用于指示在不同业务场景上不同的倾向性权重；推理大模型根据综合评分向量进行推理，输出客户的意图；根据客户的意图驱动执行相应的业务逻辑，相应的业务逻辑包括提供定制化建议或触发特定服务流程。 A method for identifying intent based on a large language model and a customer portrait classification model, the method comprising: obtaining the customer's current relevant orders and background information based on customer ID information; the customer portrait classification model obtaining a first scoring vector based on the relevant orders and background information; the first scoring vector being used to indicate the current customer portrait classification score; fine-tuning the large model to perform semantic recognition on the context of the input customer conversation to obtain a second scoring vector; the second scoring vector being used to indicate the customer's current emotional feedback score; the fine-tuning large model is obtained by fine-tuning the large language model; fusing the first scoring vector and the second scoring vector to obtain a comprehensive scoring vector; the comprehensive scoring vector being used to indicate different tendency weights in different business scenarios; the inference large model performing inference based on the comprehensive scoring vector to output the customer's intent; and driving the execution of corresponding business logic based on the customer's intent, the corresponding business logic including providing customized suggestions or triggering specific service processes.","assignee":"Hangzhou Eastcom Software Technology Co ltd","inventors":["廖翀云"],"publication_date":"2025-10-10","filing_date":"2024-07-15","priority_date":"2024-07-15","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/04","G06N5/041","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/251","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN118940838B/en"},{"publication_number":"CN120767001A","title":"A clinical trial data anomaly detection method and system based on machine learning","abstract":"本发明涉及临床数据检测技术领域，本发明涉及一种基于机器学习的临床试验数据异常检测方法及系统，其中的方法包括：构建临床试验异构图；将各节点的原始多模态特征输入到预设的混合特征编码器以得到各节点的初始向量；根据一组正常临床试验数据训练时空图谱自编码器和贝叶斯网络；通过训练好的时空图谱自编码器计算各节点的初始向量在当前时刻的重构误差向量和时空上下文嵌入；所述判定节点是否异常。本发明能够精确识别出隐藏在复杂背景下的微小偏差，提高异常检测准确率的同时，有效降低了因数据正常波动引起的误报和漏报，整体上增强了临床试验数据质量监控的可靠性与效率。 The present invention relates to the field of clinical data detection technology, and particularly to a method and system for detecting anomalies in clinical trial data based on machine learning, wherein the method comprises: constructing a clinical trial heterogeneous graph; inputting the original multimodal features of each node into a preset hybrid feature encoder to obtain an initial vector for each node; training a spatiotemporal graph autoencoder and a Bayesian network based on a set of normal clinical trial data; calculating the reconstruction error vector and spatiotemporal context embedding of the initial vector of each node at the current moment through the trained spatiotemporal graph autoencoder; and determining whether a node is abnormal. The present invention can accurately identify subtle deviations hidden in complex backgrounds, improve the accuracy of anomaly detection, and effectively reduce false positives and false negatives caused by normal data fluctuations, thereby enhancing the reliability and efficiency of clinical trial data quality monitoring as a whole.","assignee":"Yidixi Pharmaceutical Technology Jiaxing Co ltd","inventors":["袁自成"],"publication_date":"2025-10-10","filing_date":"2025-09-08","priority_date":"2025-09-08","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/70","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2415","G06F18/24155","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/243","G06F18/2433","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06F","G06F18/00","G06F18/20","G06F18/29","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06F","G06F2123/00","G06F2123/02"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120767001A/en"},{"publication_number":"CN120764583A","title":"A method for calculating the parameters of a single-layer perceptron neural network using linear programming","abstract":"本发明公开了一种基于线性规划方法的单层感知机神经网络训练方法，其实现步骤包括：从数据集中依据预测的类别个数按顺序选取样本，选取样本的方法可以是同一类别或者不同类别类一个样本。在在选取到的样本中，构造线性规划所需的约束条件参数和约束条件值，并求解线性规划。将线性规划的解赋值为单层感知机神经网络的参数。重复线性规划构造和求解过程，直至使用完所有的数据集并得到多个单层感知机。进行推理任务时，使用上述多个单层感知机进行决策，并投票选择最多的答案为最终答案。本发明基于线性规划方法求解单层感知机神经网络参数，实现单层感知机神经网络训练和预测。 The present invention discloses a single-layer perceptron neural network training method based on a linear programming method, and its implementation steps include: selecting samples in order from a data set according to the predicted number of categories, and the method of selecting samples can be one sample of the same category or different categories. In the selected samples, the constraint parameters and constraint values required for linear programming are constructed, and the linear programming is solved. The solution of the linear programming is assigned as the parameters of the single-layer perceptron neural network. The linear programming construction and solution process is repeated until all data sets are used and multiple single-layer perceptrons are obtained. When performing reasoning tasks, the above-mentioned multiple single-layer perceptrons are used to make decisions, and the answer with the most votes is the final answer. The present invention solves the single-layer perceptron neural network parameters based on the linear programming method to realize single-layer perceptron neural network training and prediction.","assignee":"Nanjing University of Posts and Telecommunications","inventors":["邓杰","夏文超"],"publication_date":"2025-10-10","filing_date":"2025-06-26","priority_date":"2025-06-26","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120764583A/en"},{"publication_number":"CN119252054B","title":"Traffic signal control method based on multi-agent deep reinforcement learning","abstract":"本发明涉及基于多智能体深度强化学习的交通信号控制方法。该方法包括：为路口设置智能体，在云端或服务器部署拟合网络，根据交通路网结构和历史交通流数据进行预测并得到交通流预测结果Y；利用该交通流预测结果和随机获取的各个路口实时交通信息，对各个智能体和拟合网络进行训练，得到各个优化后的智能体和优化后的拟合网络；各个优化后的智能体根据交通流预测结果Y和各自当前的实时交通信息进行独立观测并将其结果传输给优化后的拟合网络，计算全局奖励并反馈给各个智能体，各个智能体根据当前各自观测结果和奖励，执行下一个动作，形成目标区域当前最优的交通信号控制策略。本发明实现了多路口的交通信号协调控制，明显提升出行效率。 The present invention relates to a traffic signal control method based on multi-agent deep reinforcement learning. The method comprises: setting up an agent at an intersection, deploying a fitting network on the cloud or server, performing predictions based on the traffic network structure and historical traffic flow data, and obtaining a traffic flow prediction result Y; using the traffic flow prediction results and randomly acquired real-time traffic information from each intersection, training each agent and the fitting network to obtain optimized agents and optimized fitting networks; each optimized agent independently observes the traffic flow prediction result Y and its current real-time traffic information, and transmits the results to the optimized fitting network; calculating a global reward and feeding it back to each agent; each agent then executes the next action based on its current observation results and rewards, thereby forming the optimal traffic signal control strategy for the target area. The present invention achieves coordinated traffic signal control at multiple intersections, significantly improving travel efficiency.","assignee":"Chongqing University of Post and Telecommunications","inventors":["林峰","何帅","邵琅","蒋建春"],"publication_date":"2025-10-10","filing_date":"2024-10-25","priority_date":"2024-10-25","cpc_codes":["G","G08","G08G","G08G1/00","G08G1/07","G08G1/081","G08G1/083","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G08","G08G","G08G1/00","G08G1/01","G08G1/0104","G08G1/0125","G08G1/0129","G","G08","G08G","G08G1/00","G08G1/01","G08G1/0104","G08G1/0125","G08G1/0133","Y","Y02","Y02T","Y02T10/00","Y02T10/10","Y02T10/40"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN119252054B/en"},{"publication_number":"AU2025234251A1","title":"Infringement detection method, device and system","abstract":"A B S T R A C T An infringement detection system, comprising: one or more cameras; one or more sensors configured to detect an actively-operated vehicle and trigger the one or more cameras configured to capture, prior to any infringement detection, one or more images of at least a part of the vehicle when the actively-operated vehicle reaches an image capture point; one or more flashes configured to illuminate the vehicle or a part thereof with light at a narrow band when the actively-operated vehicle reaches the image capture point, wherein at least one of the one or more cameras comprises a narrow band filter that lets through only or substantially only wavelengths of the light produced by the one or more flashes and eliminates a majority of ambient light and/or light produced by the sun, and wherein the light produced by the one or more flashes is able to penetrate a windshield of the actively-operated vehicle; and one or more computer processors configured to automatically analyse the one or more captured images and/or one or more further images derived from the one or more captured images to detect one or more infringements corresponding to one or more of: (i) distracted driving; (ii) illegal mobile phone or mobile device use; (iii) failure to wear a seatbelt; (iv) incorrect seating; and (v) incorrect restraint; wherein, for each of the infringements, the automatic analysis comprises processing each of the one or more analysed images to generate a corresponding confidence score representing a likelihood that the analysed image shows the infringement, and using the confidence score to classify the analysed image as showing the infringement or as not showing the infringement. An infringement detection system, comprising: one or more cameras; one or more sensors configured to detect an actively-operated vehicle and trigger the one or more cameras configured to capture, prior to any infringement detection, one or more images of at least a part of the vehicle when the actively-operated vehicle reaches an image capture point; one or more flashes configured to illuminate the vehicle or a part thereof with light at a narrow band when the actively-operated vehicle reaches the image capture point, wherein at least one of the one or more cameras comprises a narrow band filter that lets through only or substantially only wavelengths of the light produced by the one or more flashes and eliminates a majority of ambient light and/or light produced by the sun, and wherein the light produced by the one or more flashes is able to penetrate a windshield of the actively-operated vehicle; and one or more computer processors configured to automatically analyse the one or more captured images and/or one or more further images derived from the one or more captured images to detect one or more infringements corresponding to one or more of: (i) distracted driving; (ii) illegal mobile phone or mobile device use; (iii) failure to wear a seatbelt; (iv) incorrect seating; and (v) incorrect restraint; wherein, for each of the infringements, the automatic analysis comprises processing each of the one or more analysed images to generate a corresponding confidence score representing a likelihood that the analysed image shows the infringement, and using the confidence score to classify the analysed image as showing the infringement or as not showing the infringement. 20 25 23 42 51 19 S ep 2 02 5 A B S T R A C T A n i n f r i n g e m e n t d e t e c t i o n s y s t e m , c o m p r i s i n g : 2 0 2 5 2 3 4 2 5 1 1 9 S e p 2 0 2 5 o n e o r m o r e c a m e r a s ; o n e o r m o r e s e n s o r s c o n f i g u r e d t o d e t e c t a n a c t i v e l y - o p e r a t e d v e h i c l e a n d t r i g g e r t h e o n e o r m o r e c a m e r a s c o n f i g u r e d t o c a p t u r e , p r i o r t o a n y i n f r i n g e m e n t d e t e c t i o n , o n e o r m o r e i m a g e s o f a t l e a s t a p a r t o f t h e v e h i c l e w h e n t h e a c t i v e l y - o p e r a t e d v e h i c l e r e a c h e s a n i m a g e c a p t u r e p o i n t ; o n e o r m o r e f l a s h e s c o n f i g u r e d t o i l l u m i n a t e t h e v e h i c l e o r a p a r t t h e r e o f w i t h l i g h t a t a n a r r o w b a n d w h e n t h e a c t i v e l y - o p e r a t e d v e h i c l e r e a c h e s t h e i m a g e c a p t u r e p o i n t , w h e r e i n a t l e a s t o n e o f t h e o n e o r m o r e c a m e r a s c o m p r i s e s a n a r r o w b a n d f i l t e r t h a t l e t s t h r o u g h o n l y o r s u b s t a n t i a l l y o n l y w a v e l e n g t h s o f t h e l i g h t p r o d u c e d b y t h e o n e o r m o r e f l a s h e s a n d e l i m i n a t e s a m a j o r i t y o f a m b i e n t l i g h t a n d / o r l i g h t p r o d u c e d b y t h e s u n , a n d w h e r e i n t h e l i g h t p r o d u c e d b y t h e o n e o r m o r e f l a s h e s i s a b l e t o p e n e t r a t e a w i n d s h i e l d o f t h e a c t i v e l y - o p e r a t e d v e h i c l e ; a n d o n e o r m o r e c o m p u t e r p r o c e s s o r s c o n f i g u r e d t o a u t o m a t i c a l l y a n a l y s e t h e o n e o r m o r e c a p t u r e d i m a g e s a n d / o r o n e o r m o r e f u r t h e r i m a g e s d e r i v e d f r o m t h e o n e o r m o r e c a p t u r e d i m a g e s t o d e t e c t o n e o r m o r e i n f r i n g e m e n t s c o r r e s p o n d i n g t o o n e o r m o r e o f : ( i ) d i s t r a c t e d d r i v i n g ; ( i i ) i l l e g a l m o b i l e p h o n e o r m o b i l e d e v i c e u s e ; ( i i i ) f a i l u r e t o w e a r a s e a t b e l t ; ( i v ) i n c o r r e c t s e a t i n g ; a n d ( v ) i n c o r r e c t r e s t r a i n t ; w h e r e i n , f o r e a c h o f t h e i n f r i n g e m e n t s , t h e a u t o m a t i c a n a l y s i s c o m p r i s e s p r o c e s s i n g e a c h o f t h e o n e o r m o r e a n a l y s e d i m a g e s t o g e n e r a t e a c o r r e s p o n d i n g c o n f i d e n c e s c o r e r e p r e s e n t i n g a l i k e l i h o o d t h a t t h e a n a l y s e d i m a g e s h o w s t h e i n f r i n g e m e n t , a n d u s i n g t h e c o n f i d e n c e s c o r e t o c l a s s i f y t h e a n a l y s e d i m a g e a s s h o w i n g t h e i n f r i n g e m e n t o r a s n o t s h o w i n g t h e i n f r i n g e m e n t .","assignee":"Acusensus IP Pty Ltd","inventors":["Alexander JANNINK"],"publication_date":"2025-10-09","filing_date":"2025-09-19","priority_date":"2018-07-19","cpc_codes":["G","G08","G08G","G08G1/00","G08G1/01","G08G1/017","G08G1/0175","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06V","G06V10/00","G06V10/40","G06V10/60","G","G06","G06V","G06V20/00","G06V20/50","G06V20/52","G06V20/54","G","G08","G08G","G08G1/00","G08G1/01","G08G1/052","G08G1/054","H","H04","H04N","H04N23/00","H04N23/20","G","G06","G06N","G06N3/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06V","G06V20/00","G06V20/60","G06V20/62","G06V20/625","G","G06","G06V","G06V2201/00","G06V2201/08","H","H04","H04L","H04L63/00","H04L63/04","H04L63/0428","H04L63/0435"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025234251A1/en"},{"publication_number":"AU2025234164A1","title":"Data processing system for generating predictions of cognitive outcome in patients","abstract":"A system for outputting a visual representation of a brain of a patient is configured to receive sensor data representing a behavior of a region of the brain of the patient. The system retrieves mapping data that maps a prediction value to the region. The prediction value is indicative of an effect on a behavior of the patient responsive to a treatment of the region, the mapping data being indexed to a patient identifier. The system receives, responsive to an application of a stimulation to the region, sensor data representing behavior of the region. The system executes a model that updates, based on the sensor data, the prediction value for the region. The system updates, responsive to executing the model, the mapping data by including the updated prediction value in the mapping data. The system outputs a visual representation of the updated mapping data comprising the updated prediction value.","assignee":"Carnegie Mellon University","inventors":["Hugo ANGULO-ORQUERA","Benjamin CHERNOFF","Bradford MAHON","Keith Parkins","Max SIMS"],"publication_date":"2025-10-09","filing_date":"2025-09-17","priority_date":"2018-11-30","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/70","A","A61","A61N","A61N1/00","A61N1/02","A61N1/04","A61N1/05","A61N1/0526","A61N1/0529","A","A61","A61B","A61B34/00","A61B34/10","A","A61","A61B","A61B5/00","A61B5/0033","A61B5/004","A61B5/0042","A","A61","A61B","A61B5/00","A61B5/0059","A61B5/0077","A","A61","A61B","A61B5/00","A61B5/05","A61B5/055","A","A61","A61B","A61B5/00","A61B5/24","A","A61","A61B","A61B5/00","A61B5/24","A61B5/316","A61B5/369","A61B5/372","A","A61","A61B","A61B5/00","A61B5/40","A61B5/4058","A61B5/4064","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4848","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A61B5/7267","A","A61","A61B","A61B5/00","A61B5/74","A61B5/742","A61B5/7425","A","A61","A61N","A61N1/00","A61N1/18","A61N1/32","A61N1/36","A61N1/36014","A","A61","A61N","A61N1/00","A61N1/18","A61N1/32","A61N1/36","A61N1/362","A61N1/37","G","G01","G01R","G01R33/00","G01R33/20","G01R33/44","G01R33/48","G01R33/4806","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H20/00","G16H20/30","G"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025234164A1/en"},{"publication_number":"US20250315659A1","title":"Embedding convolutional neural network onto integrated circuit device","abstract":"A convolutional neural network may be embedded onto an integrated circuit (IC) device, which includes an embedder unit, a flow control unit, and etched mind unit(s). The embedder unit may generate a feature map from an input image. The etched mind unit(s) may be a hardware implementation of the CNN and execute neural network operations of the CNN using the feature map. An etched mind unit may include a convolution unit implementing convolution, a batch-norm unit implementing batch normalization, an activator unit implementing an activation function operation, a max pooling unit implementing max pooling, and an average pooling unit implementing average pooling, and a MatMul unit implementing matrix multiplication, each of which may has its own memory that stores weights or other data for performing a neural network operation. The flow contour unit may orchestrate the other components of the IC device based on a timing sequence of the network.","assignee":"Individual","inventors":["Yaron Klein","Guy Yechezkel Azov","Yoni Elron","Yuval Vered"],"publication_date":"2025-10-09","filing_date":"2025-06-24","priority_date":"2024-10-17","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250315659A1/en"},{"publication_number":"US20250316046A1","title":"Oral language translation for interactivity in virtualized worlds","abstract":"Methods, systems, and computer-readable storage media are disclosed for translating a user input to a virtual environment into a contextualized output. The input is converted into a first textual representation by a recognition model, and a first set of tokens based on the textual representation is generated. The first set of tokens is fused with a second set of tokens stored in a contextualized language database. The second set of tokens is based on a second textual representation of previously collected user interactivity metrics, a virtual environment engine configuration, or displayable attributes. A trained neural network uses the fused set of tokens and at least a portion of the second set of tokens to generate an assessment of user activity to adjust a first display attribute, change the current position of the user within the virtual environment, or generate a natural language audio or textual output from the virtual environment.","assignee":"Individual","inventors":["Kenneth La-Verne Woodard, JR."],"publication_date":"2025-10-09","filing_date":"2025-06-24","priority_date":"2023-03-01","cpc_codes":["G","G06","G06T","G06T19/00","G06T19/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06F","G06F3/00","G06F3/01","G06F3/011","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0481","G06F3/04815","G","G06","G06F","G06F3/00","G06F3/01","G06F3/048","G06F3/0484","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06N","G06N5/00","G06N5/02","G","G06","G06N","G06N5/00","G06N5/04","G","G06","G06T","G06T19/00","G06T19/003","G","G06","G06T","G06T7/00","G06T7/70","G","G06","G06T","G06T2219/00","G06T2219/20","G06T2219/2004"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250316046A1/en"},{"publication_number":"US20250315658A1","title":"Neural processing unit with systolic array structure and method of operation thereof","abstract":"An operating method for a neural processing unit (NPU) is disclosed. The method includes determining that a first convolution layer performs a transpose convolution operation, dividing a kernel used for the transpose convolution into multiple sub-kernels, and performing convolution operations between an input feature map and each sub-kernel. The operations are performed by a plurality of processing elements (PEs), each configured to reuse at least one of an output feature map, a sub-kernel, or the input feature map stored in a local memory. The NPU includes a systolic array structure comprising multiple structures arranged in parallel, each corresponding to values stored in local memory. The stored values are used in successive convolution operations, enhancing computational efficiency and memory reuse in transpose convolution layers.","assignee":"DeepX Co Ltd","inventors":["Jung Boo PARK"],"publication_date":"2025-10-09","filing_date":"2025-06-24","priority_date":"2022-07-08","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06F","G06F17/00","G06F17/10","G06F17/15","G","G06","G06F","G06F17/00","G06F17/10","G06F17/15","G06F17/153","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06T","G06T1/00","G06T1/20"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250315658A1/en"},{"publication_number":"US20250313437A1","title":"Intelligent vehicle lift network with distributed sensors","abstract":"An automation system uses cameras and sensors to identify a set of vehicle wheels within a lift area, then virtualizes the position and orientation of the vehicle relative to the lift posts. Lift arms extending from the lift posts are moved to a first position that places a profile camera near the predicted position of lift points, and image analysis is used to positively identify the lift points. The lift arms are then moved to a second position that places an adapter camera below the identified lift points, and image analysis is used to confirm safe positioning below the lift point. Captured images and feedback from image analysis are used to improve and refine the system's ability to identify wheels, profile view lift points, and plan view lift points. The system may be integrated with lift arms allowing for automated rotation, extension, and elevation of lift adapters.","assignee":"Vehicle Service Group LLC","inventors":["Steven H. Taylor","Robert William Elliott","Gerry Lauderbaugh"],"publication_date":"2025-10-09","filing_date":"2025-06-23","priority_date":"2018-11-09","cpc_codes":["B","B66","B66F","B66F7/00","B66F7/28","B","B66","B66F","B66F17/00","B","B66","B66F","B66F7/00","B","B66","B66F","B66F7/00","B66F7/10","B","B66","B66F","B66F7/00","B66F7/10","B66F7/16","B66F7/20","G","G06","G06N","G06N20/00","G","G06","G06T","G06T7/00","G06T7/70","G","G06","G06V","G06V20/00","G06V20/10","H","H04","H04L","H04L67/00","H04L67/01","H04L67/12","H","H04","H04N","H04N7/00","H04N7/18","H04N7/181","H","H04","H04W","H04W4/00","H04W4/80","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30248","G06T2207/30252","G","G06","G06V","G06V2201/00","G06V2201/08"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250313437A1/en"},{"publication_number":"US20250315646A1","title":"Predictive modeling for dependency configuration in knowledge-augmented neural networks","abstract":"Data-dependent node-to-node knowledge sharing to increase the interpretability of the activation pattern of one or more nodes in a neural network, is implemented by a set of knowledge sharing links. Each link may comprise a knowledge providing node or other source P and a knowledge receiving node R. A knowledge sharing link can impose a node-specific regularization on the knowledge receiving node R to help guide the knowledge receiving node R to have an activation pattern that is more easily interpreted. The specification and training of the knowledge sharing links may be controlled by a cooperative human-AI learning supervisor system in which a human and an artificial intelligence system work cooperatively to improve the interpretability and performance of the client system.","assignee":"D5AI LLC","inventors":["James K. Baker","Bradley J. Baker"],"publication_date":"2025-10-09","filing_date":"2025-06-16","priority_date":"2020-03-23","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06F","G06F18/00","G06F18/40","G06F18/41","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250315646A1/en"},{"publication_number":"US20250317739A1","title":"Deep learning-based wireless intrusion detection","abstract":"Systems, devices, and methods for wireless intrusion detection based on deep learning are provided. A network device collects legitimate network traffic over a time period and learns a first set of features that represents the legitimate network traffic. The network device generates synthetic network traffic based on the learned first set of features and trains a machine learning model based on the learned first set of features and the synthetic network traffic. Based on the training, the machine learning model learns a second set of features that differentiates the synthetic network traffic from the legitimate network traffic. The devices and methods precisely detect potential security threats, while reducing false positives, thereby ensuring a sensitive and accurate response to genuine anomalies. Further, the devices and methods improve accuracy of detection of potential security threats including known and new attacks in wireless networks, while adapting to evolving attack techniques and network dynamics.","assignee":"Cisco Technology Inc","inventors":["Peiman Amini","Jerome Henry","Niloo BAHADORI","Bahador Amiri","Ardalan Alizadeh"],"publication_date":"2025-10-09","filing_date":"2025-02-11","priority_date":"2024-04-03","cpc_codes":["H","H04","H04W","H04W12/00","H04W12/12","H04W12/121","G","G06","G06N","G06N20/00"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250317739A1/en"},{"publication_number":"WO2025212816A1","title":"Adaptive network traffic classification","abstract":"Devices and methods for adaptively classifying network traffic associated with a new application are provided. A network device, for example, an edge device, stores a Machine Learning (ML) model pre-trained based on historical network traffic associated with a set of applications. The network device receives network traffic associated with a new application, for example, a zero-day application, that is different from the set of applications. The ML model learns one or more patterns associated with the received network traffic. The ML model detects whether the learned pattern(s) is similar to previously learned patterns of at least one application. The ML model classifies the received network traffic as legitimate traffic or anomalous traffic based on the detection. The ML model is scalable, providing timely classifications for different types of network traffic, while handling protocol and application diversity, variability in traffic patterns, and emergence of zero-day application traffic.","assignee":"Cisco Technology Inc","inventors":["Peiman Amini","Niloo Bahadori","Jerome Henry","Bahador Amiri"],"publication_date":"2025-10-09","filing_date":"2025-04-02","priority_date":"2024-04-03","cpc_codes":["H","H04","H04L","H04L63/00","H04L63/14","H04L63/1408","H04L63/1425","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","H","H04","H04L","H04L41/00","H04L41/16","H","H04","H04L","H04L43/00","H04L43/02","H04L43/026","H","H04","H04L","H04L63/00","H04L63/14","H04L63/1408","H04L63/1416","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"WO","kind":"application","source_url":"https://patents.google.com/patent/WO2025212816A1/en"},{"publication_number":"US20250315306A1","title":"Task-Based Distributional Semantic Model or Embeddings for Inferring Intent Similarity","abstract":"A course of action (CoA) monitoring system comprises a sensor and a computing system. The sensor is configured to monitor tasks included in a course of action (CoA) performed by a human operator in an environment. The computing system is in signal communication with the sensor. The computing system includes a database that stores a plurality of reference CoAs defined by reference tasks having an intended target goal, and stores a trained task-based distributional semantic model configured to determine an intent similarity of the operator performing the tasks included in the CoA during real-time. The computing system inputs the monitored tasks determined by the sensor into the trained task-based distributional semantic model to determine a deviation between the reference tasks and the monitored tasks.","assignee":"Hamilton Sundstrand Space System International Inc","inventors":["Peggy Wu"],"publication_date":"2025-10-09","filing_date":"2023-12-06","priority_date":"2023-12-06","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/20","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5027","G06F9/5038","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/20","G","G09","G09B","G09B25/00","G09B25/02"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250315306A1/en"},{"publication_number":"EP4627417A1","title":"Thermodynamic artificial intelligence for generative diffusion models and bayesian deep learning","abstract":"A physics-based system performs generative modeling on a given dataset by turning the diffusion process in diffusion models into a physical process. Electrical circuits provide an exemplary implementation, where each unit cell (an electrical circuit in a network of electrical circuits) is composed of resistors, a stochastic noise source (such as a thermal or shot noise source), programmable voltage sources, and a capacitor whose charge encodes the state variable. These unit cells can be capacitively coupled with a connectivity that matches the problem geometry. A score network, which provides predictions for score values, can be implemented on a digital, analog, or hybrid digital-arialog device. A detailed construction for an analog score network is provided in the form of a physical system that evolves over time, simultaneously with the diffusion process. The analog score network allows score values to be provided continuously to the reverse diffusion process without latency, and also allows for efficient evaluation of the loss function when coupled to the forward diffusion process.","assignee":"Normal Computing Corp","inventors":["Patrick COLES","Collin SZCZEPANSKI","Kaelan DONATELLA","Denis MELANSON","Faris SBAHI","Antonio Martinez"],"publication_date":"2025-10-08","filing_date":"2023-11-30","priority_date":"2022-12-02","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06F","G06F17/00","G06F17/10","G06F17/11","G06F17/13","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G06N3/065","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048"],"country":"EP","kind":"application","source_url":"https://patents.google.com/patent/EP4627417A1/en"},{"publication_number":"JP2025148593A","title":"Efficient Streaming Non-Recurrent On-Device End-to-End Model","abstract":"To provide an Automatic Speech Recognition (ASR) system.SOLUTION: An ASR system (109) includes a first encoder (210) configured to receive a sequence of acoustic frames (110) and generate a first high-level feature representation for corresponding acoustic frames within the sequence. The ASR system further includes a second encoder (220) configured to receive the first high-level feature representation generated by the first encoder at each of multiple output steps and to generate a second high-level feature representation for the corresponding first high-level feature frame. The ASR system further includes a decoder (204) configured to receive a second high-level feature representation generated by the second encoder at each of the multiple output steps and to generate a first probability distribution over possible speech recognition hypotheses. The ASR system further includes a language model (206) configured to receive the first probability distribution over possible speech recognition hypotheses and to generate a re-scored probability distribution (120).SELECTED DRAWING: Figure 2A","assignee":"Google LLC","inventors":["タラ・サイナス","Sainath Tara","アルン・ナラヤナン","Narayanan Arun","ラミ・ボトロス","Botros Rami","ヤンジャン・ヘ","Yanzhang He","エーサン・ヴァリアニ","Variani Ehsan","シリル・アラウゼン","Allauzen Cyril","デイヴィッド・リーバッハ","Rybach David","ルオミン・パン","Ruoming Pang","トレヴァー・ストローマン","Strohman Trevor"],"publication_date":"2025-10-07","filing_date":"2025-07-24","priority_date":"2021-03-23","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G10","G10L","G10L15/00","G10L15/02","G","G10","G10L","G10L15/00","G10L15/06","G10L15/063","G","G10","G10L","G10L15/00","G10L15/08","G10L15/16","G","G10","G10L","G10L15/00","G10L15/22","G","G10","G10L","G10L15/00","G10L15/28","G10L15/30","G","G10","G10L","G10L15/00","G10L15/28","G10L15/32"],"country":"JP","kind":"application","source_url":"https://patents.google.com/patent/JP2025148593A/en"},{"publication_number":"JP2025148542A","title":"Method for training a search result sorting model, search result sorting method, search result sorting device for training a search result sorting model, search result sorting device, electronic device, computer-readable storage medium, and computer program","abstract":"To provide a training method of a search result sorting model and a search result sorting method.SOLUTION: A method includes steps of: obtaining a plurality of first data pairs, single target features in which search results in the plurality of first data pairs correspond to a plurality of search targets, and an annotation score in which the plurality of first data pairs respectively correspond to the plurality of search targets; obtaining a plurality of single search target sorting models by training based on the obtained data; obtaining a plurality of second data pairs and multi-target features in which respective search results in the plurality of second data pairs correspond to all search targets; and scoring the respective search results in the second data pairs using the plurality of single search target sorting models corresponding to the plurality of search targets to determine training search targets of the second data pairs; and obtaining the search result sorting models by training based on queries in the plurality of second data pairs, the multi-target features for which the respective search results correspond to all search targets, and scores for which the respective search results correspond to the training search targets.SELECTED DRAWING: Figure 1","assignee":"Beijing Baidu Netcom Science and Technology Co Ltd","inventors":["ワン、ハイフェン","Haifeng Wang","ティアン、ハオ","Hao Tian","ウ、フア","Hua Wu","ウ、ティアン","Tian Wu","リウ、ジン","Jing Liu","チェン、ウェイジェン","Weizheng Chen","ダイ、ダイ","Dai Dai","ワン、ジアカン","Jiakang Wang","パン、チャオ","Chao Pang","ワン、ウェンフア","Wenhua Wang"],"publication_date":"2025-10-07","filing_date":"2025-07-17","priority_date":"2022-06-27","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G","G06","G06F","G06F16/00","G06F16/30","G06F16/38","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","Y","Y02","Y02D","Y02D10/00"],"country":"JP","kind":"application","source_url":"https://patents.google.com/patent/JP2025148542A/en"},{"publication_number":"JP2025148425A","title":"Semiconductor Devices","abstract":"To provide a semiconductor device which can improve arithmetic capacity such as AI learning and processing efficiency.SOLUTION: A flash memory 100 includes a NAND type or NOR type memory cell array 110, and an arithmetic processing unit 190. The arithmetic processing unit 190 includes a bit line current detection unit 200, a voltage holding unit 210 which holds voltage corresponding to detected current, an addition unit 220 which adds the voltage held in the voltage holding unit 210, and an A/D conversion unit 230 which performs A/D conversion on an addition result of the addition unit 220. The arithmetic processing unit 190 can calculate the sum of currents which flow through a bit line in a line direction and/or a column direction at reading of the memory cell array.SELECTED DRAWING: Figure 2","assignee":"Winbond Electronics Corp","inventors":["勝 矢野","Masaru Yano"],"publication_date":"2025-10-07","filing_date":"2025-07-08","priority_date":"2023-04-12","cpc_codes":["G","G11","G11C","G11C16/00","G11C16/02","G11C16/06","G11C16/30","G","G11","G11C","G11C11/00","G11C11/56","G11C11/5621","G11C11/5642","G","G06","G06N","G06N3/00","G06N3/02","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G11","G11C","G11C11/00","G11C11/54","G","G11","G11C","G11C11/00","G11C11/56","G11C11/5621","G11C11/5628","G","G11","G11C","G11C16/00","G11C16/02","G11C16/04","G11C16/0483","G","G11","G11C","G11C16/00","G11C16/02","G11C16/06","G11C16/10","G","G11","G11C","G11C16/00","G11C16/02","G11C16/06","G11C16/26"],"country":"JP","kind":"application","source_url":"https://patents.google.com/patent/JP2025148425A/en"},{"publication_number":"JP2025148395A","title":"Systems and methods for quantum computing","abstract":"【課題】ホスト材料中に含有されるドーパント分子を量子ビットとして利用する、非古典的（例えば、量子）計算システムおよび方法に関する。 【解決手段】本開示は、ホスト材料中に含有されるドーパント分子を量子ビットとして利用する、非古典的（例えば、量子）計算システムおよび方法を記載する。 【選択図】図１Ａ The present invention relates to non-classical (eg, quantum) computing systems and methods that utilize dopant molecules contained in a host material as qubits. The present disclosure describes non-classical (eg, quantum) computing systems and methods that utilize dopant molecules contained in a host material as qubits. [Selected Figure] Figure 1A","assignee":"Nvision Imaging Technologies GmbH","inventors":["シュバルツ イライ","Schwartz Ilai","プフェンダー マティーアス","Pfender Matthias","シャウプ トビアス","Schaub Tobias","ノイマン フィリップ","Philipp Neumann"],"publication_date":"2025-10-07","filing_date":"2025-07-03","priority_date":"2021-05-12","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/40","B","B82","B82Y","B82Y10/00","H","H10","H10N","H10N60/00","H10N60/10","H10N60/12"],"country":"JP","kind":"application","source_url":"https://patents.google.com/patent/JP2025148395A/en"},{"publication_number":"US12437190B2","title":"Automated fine-tuning of a pre-trained neural network for transfer learning","abstract":"In an embodiment, a method for fine-tuning a pre-trained neural network for transfer learning, the method comprising obtaining a first target feature vector from a first layer of a pre-trained neural network responsive to a first target data element of a target dataset passing therethrough, obtaining a first source feature vector associated with the first layer of the pre-trained neural network, calculating a first divergence value for the first layer of the pre-trained neural network based at least in part on the first target feature vector and the first source feature vector, and setting a learning rate for the first layer of the pre-trained neural network based at least in part on the first divergence value.","assignee":"International Business Machines Corp","inventors":["Parijat Dube","Bishwaranjan Bhattacharjee","Patrick Watson","John Ronald Kender"],"publication_date":"2025-10-07","filing_date":"2019-12-05","priority_date":"2019-12-05","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06F","G06F17/00","G06F17/10","G06F17/18","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12437190B2/en"},{"publication_number":"US12433511B2","title":"Systems, methods, and devices for biophysical modeling and response prediction","abstract":"Various systems and methods are disclosed. One or more of the methods disclosed uses machine learning algorithms to predict biophysical responses from biophysical data, such as heart rate monitor data, food logs, or glucose measurements. Biophysical responses may include behavioral responses. Additional systems and methods extract nutritional information from food items by parsing strings containing names of food items.","assignee":"January Inc","inventors":["Parin Bhadrik Dalal","Salar Rahili","Solmaz Shariat Torbaghan","Saransh Agarwal","Mehrdad Yazdani"],"publication_date":"2025-10-07","filing_date":"2020-02-07","priority_date":"2018-11-29","cpc_codes":["A","A61","A61B","A61B5/00","A61B5/0002","A61B5/0015","A61B5/0022","A","A61","A61B","A61B5/00","A61B5/0002","A61B5/0015","A61B5/0024","A","A61","A61B","A61B5/00","A61B5/02","A61B5/024","A","A61","A61B","A61B5/00","A61B5/02","A61B5/024","A61B5/02438","A","A61","A61B","A61B5/00","A61B5/145","A61B5/14532","A","A61","A61B","A61B5/00","A61B5/68","A61B5/6801","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7271","A61B5/7275","G","G06","G06F","G06F40/00","G06F40/20","G06F40/205","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/088","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H20/00","G16H20/60","G","G16","G16H","G16H40/00"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12433511B2/en"},{"publication_number":"US12437207B2","title":"Method for selecting task network, system and method for determining actions based on sensing data","abstract":"The embodiments of the disclosure provide a method for selecting a task network, a system and a method for determining actions based on sensing data. The method of the embodiments of the disclosure includes: mapping the sensing data into an input feature vector; feeding the input feature vector into a specific task network to generate an output feature vector via the specific task network, in which the specific task network is trained based on a plurality of first individuals and a plurality of second individuals, the first individuals belong to a first population, the second individuals belong to a second population, and the first individuals and the second individuals are evolved via a coevolution process; and determining an output action according to the output feature vector, and setting a second specific individual based on the output action, in which the second specific individual belongs to the second population.","assignee":"Wistron Corp","inventors":["Chih-Ming Chen"],"publication_date":"2025-10-07","filing_date":"2021-01-14","priority_date":"2020-11-04","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2134","G06F18/21342","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/217","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2411","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2413","G","G06","G06N","G06N20/00","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N3/00","G06N3/004","G06N3/006","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0495","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/086","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N3/00","G06N3/12","G06N3/126"],"country":"US","kind":"grant","source_url":"https://patents.google.com/patent/US12437207B2/en"},{"publication_number":"JP2025147225A","title":"Machine learning device, machine learning method, and machine learning program","abstract":"To provide a machine learning technique capable of performing transition learning according to characteristics of a domain.SOLUTION: When a first model of neural network learned using teacher data of a first domain is subjected to transition learning by teacher data of a second domain, a domain adaptive data fullness degree determining part 52 determines a domain adaptive data fullness degree on the basis of the number of teacher data of a second domain. A learning layer determining part 54 determines a layer of a learning object of a second model obtained by copying the first model, based on the domain adaptive data fullness degree. A transition learning execution part 56 subjects the layer of the learning object of the second model to transition learning using the teacher data of the second domain.SELECTED DRAWING: Figure 2","assignee":"JVCKenwood Corp","inventors":["英樹 竹原","Hideki Takehara","晋吾 木田","Shingo Kida","尹誠 楊","Yincheng Yang"],"publication_date":"2025-10-06","filing_date":"2025-08-07","priority_date":"2021-02-10","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096"],"country":"JP","kind":"application","source_url":"https://patents.google.com/patent/JP2025147225A/en"},{"publication_number":"LU600482B1","title":"Ein Online-Erkennungs- und -Verarbeitungsverfahren für multivariate Datenanomalien","abstract":"Die vorliegende Erfindung offenbart ein Online-Erkennungs- und Verarbeitungsverfahren für multivariate Datenanomalien, bezieht sich auf den Bereich der Echtzeit-Überwachung von Großgeräten und umfasst Schritt S1, Analyse der Erscheinungsmuster von Datenanomalien, Zusammenfassung der Datenverteilungsmerkmale unter verschiedenen Mustern und Aufteilung der Anomalieverarbeitung in zwei Kategorien von Alarmierung und Reparatur; Schritt S2, Entwurf der Beurteilungsrichtlinien für verschiedene Arten von Anomalien entsprechend den Datenverteilungsmerkmalen und Speichern der Richtlinienparameter als Datei; Schritt S3: Das Online-System lädt die Richtlinienparameter, entwirft ein einheitliches Modell zur Erkennung von Anomalien, bestimmt, ob die Daten in Echtzeit anormal sind, sendet die Informationen rechtzeitig an das zuständige Personal, wenn ein Alarm erforderlich ist, und korrigiert die Anomalie anhand der historischen Daten, wenn eine Reparatur erforderlich ist. Die vorliegende Erfindung analysiert vollständig die morphologischen Merkmale von multivariaten Datenanomalien, entwirft gezielte Beurteilungsrichtlinien, stellt ein einheitliches Anomalieerkennungsmodell für multivariate Daten auf und hilft, die Effizienz und Praktikabilität der Online-Erkennung von Datenanomalien zu verbessern.","assignee":"Nanjing Univ Of Finance & Economics","inventors":["Yadong Dou"],"publication_date":"2025-10-06","filing_date":"2025-03-06","priority_date":"2025-03-06","cpc_codes":["G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0259","G05B23/0275","G","G05","G05B","G05B19/00","G05B19/02","G05B19/04","G05B19/048","G","G06","G06F","G06F17/00","G","G06","G06N","G06N3/00"],"country":"LU","kind":"grant","source_url":"https://patents.google.com/patent/LU600482B1/en"},{"publication_number":"NL2037342B1","title":"Method for balancing unbalanced data for use in training a machine learning model","abstract":"Method for balancing unbalanced data for use in training a machine learning model, wherein data is defined to be unbalanced if said data contains a first set of data items belonging to a first set of categories as well as a second set of data items belonging to a second set of categories distinct from the first set of categories, wherein the first set of categories is more numerous than the second set of categories and the second set of data items is more numerous than the first set of data items; the method comprising: for each data item in a source batch comprising a plurality of data items of unbalanced data, categorizing said data item into a category among a plurality of categories; storing each categorized data item in a respective priority queue of a plurality of priority queues, the respective priority queue corresponding with the category associated with said categorized data item; and sampling the plurality of priority queues without replacement in accordance with a target distribution in order to generate the target batch in statistical conformity with the target distribution.","assignee":"Ai4R B V","inventors":["Eugenius Van Deursen Mitch","Lucas Verdenius Stijn","Lee Marvin Wang Roy"],"publication_date":"2025-10-06","filing_date":"2024-03-27","priority_date":"2024-03-27","cpc_codes":["G","G06","G06N","G06N20/00"],"country":"NL","kind":"grant","source_url":"https://patents.google.com/patent/NL2037342B1/en"},{"publication_number":"JP2025533350A","title":"quantum interior point method","abstract":"いくつかの態様において、本明細書で述べられる技術は、二次錐計画問題（ＳＯＣＰ）のインスタンスを解くための量子方法に関連し、ＳＯＣＰインスタンスに基づいて行列Ｇ及びベクトルｈ → を構築することにより、ＳＯＣＰインスタンスに対するニュートン系を定義すること、行列Ｇ及びベクトルｈ → を行の正規化により、行列Ｇの条件数を減少させること、所定の反復条件が満たされるまでｕ → を反復的に決定し、この反復には、量子計算システムに、行列Ｇ及びベクトルｈ → を量子線形系ソルバー（ＱＬＳＳ）に適用して、量子状態を生成させことと、量子コンピューティングシステムに、量子状態に対して量子状態トモグラフィーを実行させることと、ｕ → の現在の値及び量子状態トモグラフィーの出力に基づいて、ｕ → の値を更新すること、を含むこと、並びに、ｕ → の更新された値に基づいて、ＳＯＣＰインスタンスの解を決定すること、を含む。 In some aspects, the techniques described herein relate to quantum methods for solving instances of second-order cone programming problems (SOCP), including defining a Newtonian system for the SOCP instance by constructing a matrix G and a vector h → based on the SOCP instance; row-normalizing the matrix G and the vector h → to reduce the condition number of the matrix G; iteratively determining u → until a predetermined iteration condition is met, the iteration including having a quantum computing system apply the matrix G and the vector h → to a quantum linear system solver (QLSS) to generate a quantum state; having the quantum computing system perform quantum state tomography on the quantum state; and updating the value of u → based on the current value of u → and the output of the quantum state tomography; and determining a solution to the SOCP instance based on the updated value of u → .","assignee":"ゴールドマン サックス アンド カンパニー エルエルシー","inventors":["エム．ダルゼル アレクサンダー","デイビッド クラダー ブライアン","サルトン グラント","ベルタ マリオ","イェン－ユー リン セドリック","アルバート バーダー デイビッド","ジョセフ ゼン ウィリアム"],"publication_date":"2025-10-06","filing_date":"2023-10-04","priority_date":"2022-10-04","cpc_codes":["G","G06","G06N","G06N10/00","G06N10/20","G","G06","G06N","G06N10/00","G06N10/60"],"country":"JP","kind":"application","source_url":"https://patents.google.com/patent/JP2025533350A/en"},{"publication_number":"JP2025533342A","title":"Systems, methods, and computer devices for aggregate thresholding, adaptive cropping, and image classification for anomaly detection in machine vision applications","abstract":"視覚検査および異常検出のためのシステムおよび方法が開示される。検査画像がゴールデンサンプル画像と比較され、異常マップが特定される。当該異常マップに対して集約閾値処理が実行され、異常が識別される。識別された異常に対して適応的クロッピングが実行され、当該異常のクロップ画像が取得される。取得されたクロップ画像は、擬似ワンクラス分類器である画像分類モデルに提供される。当該画像分類モデルが、異常を分類する。 A system and method for visual inspection and anomaly detection is disclosed. An inspection image is compared to a golden sample image to identify an anomaly map. Aggregate thresholding is performed on the anomaly map to identify anomalies. Adaptive cropping is performed on the identified anomalies to obtain cropped images of the anomalies. The cropped images are provided to an image classification model, which is a pseudo-one-class classifier. The image classification model classifies the anomalies.","assignee":"ムサシ エーアイ ノース アメリカ インコーポレイテッド","inventors":["バクシュマンド、サイード"],"publication_date":"2025-10-06","filing_date":"2023-10-04","priority_date":"2022-10-04","cpc_codes":["G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0004","G06T7/001","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06T","G06T7/00","G06T7/10","G06T7/13","G","G06","G06V","G06V10/00","G06V10/20","G06V10/25","G","G06","G06V","G06V10/00","G06V10/20","G06V10/26","G","G06","G06V","G06V10/00","G06V10/20","G06V10/28","G","G06","G06V","G06V10/00","G06V10/20","G06V10/32","G","G06","G06V","G06V10/00","G06V10/20","G06V10/34","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/40","G","G06","G06V","G06V20/00","G06V20/70","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10016","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20084","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20112","G06T2207/20132","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20212","G06T2207/20224","G","G06","G06T","G06T2207/00","G06T2207/30","G06T2207/30108","G06T2207/30164","G","G06","G06T","G06T2210/00","G06T2210/12"],"country":"JP","kind":"application","source_url":"https://patents.google.com/patent/JP2025533342A/en"},{"publication_number":"JP2025533327A","title":"How servers manage restrooms based on restroom usage","abstract":"サーバーが利用者情報に基づいてトイレを管理する方法であって、前記トイレに設置された少なくとも１つのセンサーを介して利用者情報を取得する情報取得ステップと、前記利用者情報に基づいてトイレの状態を把握して少なくとも１つの端末に送信するトイレ状態送信ステップと、前記トイレ状態送信ステップで把握されたトイレの状態に応じて掃除担当者端末に掃除を要求する掃除要求ステップと、を含む。 【選択図】図１ A method for a server to manage toilets based on user information includes an information acquisition step of acquiring user information via at least one sensor installed in the toilet, a toilet status transmission step of determining the status of the toilet based on the user information and transmitting the information to at least one terminal, and a cleaning request step of requesting cleaning from a cleaning staff terminal according to the status of the toilet determined in the toilet status transmission step. [Selected Figure] Figure 1","assignee":"Uniuni Corp","inventors":["ヨン ハン、ス"],"publication_date":"2025-10-06","filing_date":"2022-12-28","priority_date":"2022-11-29","cpc_codes":["A","A47","A47K","A47K17/00","G","G01","G01D","G01D21/00","G01D21/02","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/02","G","G06","G06Q","G06Q50/00","G06Q50/10","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26","G","G08","G08B","G08B21/00","G08B21/18","H","H04","H04W","H04W4/00","H04W4/02"],"country":"JP","kind":"application","source_url":"https://patents.google.com/patent/JP2025533327A/en"},{"publication_number":"ES3037715T3","title":"Methods and systems for boosting deep neural networks for deep learning","abstract":"Se describen métodos y sistemas para potenciar redes neuronales profundas para el aprendizaje profundo. En un ejemplo, en una red neuronal profunda que incluye una primera red superficial y una segunda red superficial, la primera red superficial procesa una primera muestra de entrenamiento con pesos iguales. Se determina una pérdida para la primera red superficial a partir de la muestra de entrenamiento procesada con pesos iguales. Los pesos para la segunda red superficial se ajustan en función de la pérdida determinada para la primera red superficial. La segunda red superficial procesa una segunda muestra de entrenamiento con los pesos ajustados. En otro ejemplo, en una red neuronal profunda que incluye una primera red débil y una segunda red débil, la primera red débil procesa un primer subconjunto de muestras de entrenamiento con pesos inicializados. Se determina un error de clasificación para la primera red débil en el primer subconjunto de muestras de entrenamiento. La segunda red débil se potencia utilizando el error de clasificación determinado de la primera red débil con pesos ajustados. La segunda red débil procesa un segundo subconjunto de muestras de entrenamiento con los pesos ajustados. (Traducción automática con Google Translate, sin valor legal) Methods and systems for enhancing deep neural networks for deep learning are described. In one example, in a deep neural network comprising a first surface network and a second surface network, the first surface network processes a first training sample with equal weights. A loss for the first surface network is determined from the training sample processed with equal weights. The weights for the second surface network are adjusted based on the loss determined for the first surface network. The second surface network processes a second training sample with the adjusted weights. In another example, in a deep neural network comprising a first weak network and a second weak network, the first weak network processes a first subset of training samples with initialized weights. A classification error for the first weak network is determined on the first subset of training samples. The second weak network is enhanced using the classification error determined from the first weak network with adjusted weights. The second weak network processes a second subset of training samples with the adjusted weights.","assignee":"Intel Corp","inventors":["Libin Wang","Yiwen Guo","Anbang Yao","Dongqi Cai","Lin Xu","Ping Hu","Shandong Wang","Wenhua Cheng","Yurong Chen"],"publication_date":"2025-10-06","filing_date":"2017-04-07","priority_date":"2017-04-07","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G06F18/2148","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/217","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/096","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098","G","G06","G06T","G06T1/00","G06T1/20"],"country":"ES","kind":"application","source_url":"https://patents.google.com/patent/ES3037715T3/en"},{"publication_number":"CN120746793A","title":"Knowledge tracking method and system based on cognitive decoupling","abstract":"The invention discloses a knowledge tracking method and a knowledge tracking system based on cognitive decoupling, which belong to the technical field of knowledge tracking, wherein the method comprises the steps of encoding historical learning interaction data of students to generate a topic feature vector and a response feature vector; the method comprises the steps of respectively decomposing the question feature vector and the answer feature vector into a stable cognition mode component and a random factor component, carrying out time sequence modeling on the stable cognition mode component based on an attenuation attention mechanism, simulating short-term disturbance of the random factor component to a knowledge state based on the attenuation attention mechanism to obtain dynamic knowledge state representation, and predicting correct answer probability of students to the next question according to the dynamic knowledge state representation and the current question feature vector. The invention solves the problems of noise sensitivity, insufficient dynamic modeling, single characteristic characterization and the like in the prior art, and finally realizes higher prediction precision, stronger robustness and finer granularity knowledge state tracking.","assignee":"Jinan University","inventors":["郭腾","夏煜彬","秦宇","侯明良","刘子韬"],"publication_date":"2025-10-03","filing_date":"2025-09-08","priority_date":"2025-09-08","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/047","G","G06","G06Q","G06Q10/00","G06Q10/04"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120746793A/en"},{"publication_number":"CN120751140A","title":"End-to-end image file processing method, model construction method and device","abstract":"The invention relates to an end-to-end image file processing method, a model construction method and a device, which comprise the steps of obtaining an image file to be processed, inputting the image file into a pre-constructed image report generation model to carry out image semantic coding processing, generating a context vector and text sequence prediction decoding processing based on a space-time attention mechanism, and outputting to obtain an image report text, wherein the generation of the context vector based on the space-time attention mechanism comprises the steps of carrying out dynamic mapping learning on the corresponding relation between importance degrees of different image areas and text predictions of each time step based on the attention mechanism according to the result of the image semantic coding processing and the result of the text sequence prediction decoding processing, and obtaining the context vector of each time step for a global image, wherein the context vector is obtained after weighting attention weight according to the result of the image semantic coding processing. The output report has higher accuracy, high processing efficiency and good generalization performance of the model.","assignee":"Tongxin Intelligent Medical Technology Beijing Co ltd","inventors":["刘伟奇","仇壮","马嘉婧","李鑫","郭星含"],"publication_date":"2025-10-03","filing_date":"2025-09-08","priority_date":"2025-09-08","cpc_codes":["H","H04","H04N","H04N19/00","H04N19/10","H04N19/134","H04N19/154","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06T","G06T7/00","G06T7/0002","G06T7/0012","H","H04","H04N","H04N17/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120751140A/en"},{"publication_number":"CN120744643A","title":"Multi-mode theme classification method integrating image semantics","abstract":"本发明涉及主题分类处理技术领域，具体地说，涉及一种融合图像语义的多模态主题分类方法，用于内容审核，其步骤包括：获取含文字说明和配图的新闻内容项；提取文字特征与图片视觉特征；获得加权后的文字及图片核心特征；计算图文特征关联权重，生成图片摘要特征并与文字核心特征融合，得出表示图文语义一致程度的一致性分数；将该分数与预设阈值比较，低于阈值则判定为潜在图文误导信息，依据关联权重生成解释性报告，并高亮显示关联权重最低的图像区域与关键词。本发明通过自动量化图文一致性，精准识别误导信息，提升内容审核自动化水平和人工复核效率，对净化网络信息环境具有重要价值。 The present invention relates to the technical field of topic classification processing, specifically, to a multimodal topic classification method that integrates image semantics and is used for content review, the steps of which include: obtaining news content items containing text descriptions and accompanying pictures; extracting text features and image visual features; obtaining weighted text and image core features; calculating the association weights of image and text features, generating image summary features and integrating them with text core features to obtain a consistency score that represents the degree of semantic consistency between the image and text; comparing the score with a preset threshold, and determining that it is potentially misleading information if it is lower than the threshold, generating an explanatory report based on the association weights, and highlighting the image areas and keywords with the lowest association weights. The present invention automatically quantifies image and text consistency, accurately identifies misleading information, improves the automation level of content review and the efficiency of manual review, and is of great value in purifying the network information environment.","assignee":"Sichuan Agricultural University","inventors":["覃涵","李雅桐","刘沿汐","刘瑞","鲜磊","胡喜贵","黄春寒","王学文","陈妍琳","牛昱澎"],"publication_date":"2025-10-03","filing_date":"2025-09-08","priority_date":"2025-09-08","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G","G06","G06F","G06F18/00","G06F18/20","G06F18/27","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120744643A/en"},{"publication_number":"CN120747941A","title":"Sequence perception anti-attack method and system for scene text recognition model","abstract":"本发明提出一种针对场景文本识别模型的序列感知对抗攻击方法及系统，属于计算机视觉与对抗机器学习领域，包括：S1：获取数据集和场景文本识别模型的信息；S2：将场景文本识别模型转为测试阶段，筛选出模型能识别的输入图片；S3：构建并执行攻击算法，生成对抗样本，并构造损失函数以减小对抗样本和输入图片的差异。本发明的方法利用场景文本识别模型的序列预测特性，显著提高攻击成功率并减少扰动幅度，为场景文本识别模型的鲁棒性评估提供高效攻击工具，也能用于现在前沿的隐私保护和版权保护方案。 This paper proposes a sequence-aware adversarial attack method and system for scene text recognition models, belonging to the fields of computer vision and adversarial machine learning. The method comprises the following steps: S1: obtaining a dataset and information about the scene text recognition model; S2: switching the scene text recognition model to a testing phase to select input images that the model can recognize; and S3: constructing and executing an attack algorithm, generating adversarial samples, and constructing a loss function to minimize the discrepancy between the adversarial samples and the input images. This method leverages the sequence prediction properties of the scene text recognition model to significantly improve the attack success rate and reduce the perturbation amplitude. This method provides an efficient attack tool for robustness assessment of scene text recognition models and can also be used in cutting-edge privacy and copyright protection schemes.","assignee":"Communication University of China","inventors":["徐翊焜"],"publication_date":"2025-10-03","filing_date":"2025-09-08","priority_date":"2025-09-08","cpc_codes":["G","G06","G06V","G06V20/00","G06V20/60","G06V20/62","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/094","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V30/00","G06V30/10","G06V30/19","G06V30/191"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120747941A/en"},{"publication_number":"CN120744644A","title":"Unmanned aerial vehicle power distribution network inspection image real-time identification method based on artificial intelligence","abstract":"本发明公开了基于人工智能的无人机配电网巡检图像实时识别方法，涉及图像识别领域，包括，对多模态数据包进行时空校准，提取各模态特征后使用跨模态注意力机制融合特征，得到多模态特征向量，通过配电网拓扑数据构建异构图，设置节点和边，将多模态特征向量赋给节点，使用图卷积神经网络聚合邻居设备特征，更新节点表示，输出配电网设备的异常分类结果和异常传播路径预测结果，带宽网络中心接收异常分类结果和异常传播路径预测结果，并持续监测和利用历史网络数据与环境因素训练带宽预测模型，预测带宽变化趋势；本发明实现了对设备间复杂依赖关系的捕捉和异常传播路径的精准预测，显著提升巡检的全面性和故障预防能力。 The present invention discloses a real-time recognition method for UAV distribution network inspection images based on artificial intelligence, which relates to the field of image recognition. The method comprises the following steps: performing spatiotemporal calibration on multimodal data packets, extracting features of each modality and fusing the features using a cross-modal attention mechanism to obtain a multimodal feature vector, constructing a heterogeneous graph through distribution network topology data, setting nodes and edges, assigning multimodal feature vectors to nodes, aggregating neighboring device features using a graph convolutional neural network, updating node representations, outputting abnormal classification results and abnormal propagation path prediction results for distribution network equipment, receiving the abnormal classification results and abnormal propagation path prediction results by a bandwidth network center, and continuously monitoring and utilizing historical network data and environmental factors to train a bandwidth prediction model to predict bandwidth change trends. The present invention achieves the capture of complex dependencies between devices and the accurate prediction of abnormal propagation paths, significantly improving the comprehensiveness of inspections and fault prevention capabilities.","assignee":"Yongcheng Power Distribution Network Construction Branch Of Ningbo Power Transmission And Distribution Construction Co ltd","inventors":["崔林宁","丁钰雯","杜铮","王军华","张蔡洧","姜春莹","陈果","雷启迪","王永慧","孙圳","谌庆芳","张科","秦昊","周开泓","陈晗文","程日朗","李晨辉","许静波","徐璐","周乐平","周浩","欧竞"],"publication_date":"2025-10-03","filing_date":"2025-09-08","priority_date":"2025-09-08","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G","G01","G01D","G01D21/00","G01D21/02","G","G01","G01R","G01R31/00","G01R31/08","G01R31/081","G01R31/086","G","G01","G01R","G01R31/00","G01R31/08","G01R31/088","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06Q","G06Q50/00","G06Q50/06","G","G06","G06V","G06V10/00","G06V10/70","G06V10/764","G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/80","G06V10/806","G","G06","G06V","G06V10/00","G06V10/70","G06V10/82","G","G06","G06V","G06V20/00","G06V20/10","G06V20/17"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120744644A/en"},{"publication_number":"CN107636665B","title":"Cascade classifiers for computer security applications","abstract":"所描述的系统及方法允许计算机安全系统使用经训练分类器级联来对目标对象进行自动分类以用于包含恶意软件检测、垃圾邮件检测及/或欺诈检测在内的应用。所述级联包括数个层级，每一层级包含一组分类器。按照分类器的相应层级的预定次序来训练所述分类器。每一分类器经训练以将记录语料库划分成多个记录群组，使得一个此类群组中的所述记录的相当大比例(例如，至少95％或全部)是同一类别的成员。在训练所述级联的连续层级的分类器之间，从训练语料库摒弃相应群组的一组训练记录。当用于对未知目标对象进行分类时，一些实施例按照所述分类器的相应层级的次序采用所述分类器。 The described systems and methods allow a computer security system to automatically classify target objects using a cascade of trained classifiers for applications including malware detection, spam detection, and/or fraud detection. The cascade includes several levels, each level including a set of classifiers. The classifiers are trained in a predetermined order of their respective levels. Each classifier is trained to partition a corpus of records into a plurality of groups of records such that a substantial proportion (e.g., at least 95% or all) of the records in one such group are members of the same class. Between training classifiers in successive levels of the cascade, a set of training records for the respective group is discarded from the training corpus. When used to classify an unknown target object, some embodiments employ the classifiers in the order of their respective levels.","assignee":"Bitdefender IPR Management Ltd","inventors":["D-T·加夫里卢特","C·瓦塔马努","D·科索万","H·卢基安"],"publication_date":"2025-10-03","filing_date":"2016-05-07","priority_date":"2015-05-17","cpc_codes":["G","G06","G06F","G06F21/00","G06F21/50","G06F21/51","G","G06","G06F","G06F21/00","G06F21/50","G06F21/55","G06F21/552","G","G06","G06F","G06F21/00","G06F21/50","G06F21/55","G06F21/56","G","G06","G06F","G06F21/00","G06F21/50","G06F21/55","G06F21/56","G06F21/566","G","G06","G06N","G06N20/00","H","H04","H04L","H04L63/00","H04L63/14","H","H04","H04L","H04L63/00","H04L63/14","H04L63/1408","H04L63/1425","G","G06","G06F","G06F2221/00","G06F2221/03","G06F2221/033","G","G06","G06F","G06F2221/00","G06F2221/03","G06F2221/034"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN107636665B/en"},{"publication_number":"CN114169492B","title":"Neural networks for processing graph data","abstract":"The present invention relates to neural networks for processing graphics data. Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for receiving graphics data representing an input graphic, the input graphic including a plurality of vertices connected by edges, generating vertex input data representing characteristics of each vertex in the input graphic and pairing input data representing characteristics of pairs of vertices in the input graphic from the graphics data, and generating order invariant features of the input graphic using a neural network, wherein the neural network includes a first subnetwork configured to generate a first alternative representation of the vertex input data and a first alternative representation of the pairing input data from the vertex input data and the pairing input data, and a combining layer configured to receive an input alternative representation and process the input alternative representation to generate the order invariant features.","assignee":"Google LLC","inventors":["帕特里克·F·赖利","马克·伯恩德尔"],"publication_date":"2025-10-03","filing_date":"2016-08-12","priority_date":"2015-09-01","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/241","G06F18/2413","G06F18/24133","G06F18/24137","G06F18/2414","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G16","G16C","G16C20/00","G16C20/70"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN114169492B/en"},{"publication_number":"CN120336932B","title":"An event prediction and early warning method based on Bayesian deep learning","abstract":"The invention discloses an event prediction early warning method based on Bayes deep learning, which mainly comprises two parts, namely a Bayes deep learning rule model construction based on event characteristics and a prediction early warning model design developed on the basis of the Bayes deep learning rule model construction. According to the invention, the advantages of the Bayesian statistical method and the deep learning model are fused, the uncertainty of the prediction result can be expressed in a probability distribution form by combining prior knowledge and observation data, the complex and nonlinear time series data is efficiently modeled and predicted, the early warning of the potential event is realized, and the accuracy and reliability of the event prediction warning are improved.","assignee":"Hangzhou Maquan Information Technology Co ltd","inventors":["闫鑫怡","许永恩","王昶庆","马汉杰","王标"],"publication_date":"2025-10-03","filing_date":"2025-06-10","priority_date":"2025-06-10","cpc_codes":["G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G","G06","G06F","G06F18/00","G06F18/10","G06F18/15","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/211","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G06F18/2135","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120336932B/en"},{"publication_number":"CN120744065A","title":"Group consensus large model illusion reducing method based on multi-model challenge","abstract":"本发明涉及一种基于多模型诘问的群体共识大模型幻觉降低方法，包括：根据多维综合评分结果从候选模型集中筛选出Top‑N模型，对Top‑N模型执行全组合双向知识蒸馏，得到初始模型群体，对初始模型群体中每个模型加载多个领域知识库进行领域自适应微调，构建得到群体模型集；给定用户问题，触发群体模型集中多个微调后的领域专家模型进行并行推理生成初始回答，通过构建诘问集、生成诘问指令以及更新回答进行迭代优化，在迭代优化过程中采用混合核函数计算群体回答的相似度，当相似度和稳定性同时达到预设阈值，或达到最大迭代次数时，终止迭代并输出结果。与现有技术相比，本发明具有回答准确性高以及领域适应性强等优点。 The present invention relates to a method for reducing the illusion of a large model of group consensus based on multi-model questioning, comprising: screening out the top-N models from a candidate model set according to a multi-dimensional comprehensive scoring result, performing full-combination bidirectional knowledge distillation on the top-N models to obtain an initial model group, loading multiple domain knowledge bases on each model in the initial model group for domain adaptive fine-tuning, and constructing a group model set; given a user question, triggering multiple fine-tuned domain expert models in the group model set to perform parallel reasoning to generate an initial answer, iteratively optimizing by constructing a question set, generating questioning instructions, and updating answers, and using a hybrid kernel function to calculate the similarity of the group answers during the iterative optimization process. When the similarity and stability simultaneously reach a preset threshold, or when the maximum number of iterations is reached, the iteration is terminated and the result is output. Compared with the prior art, the present invention has the advantages of high answer accuracy and strong domain adaptability.","assignee":"Shanghai Academy of Agricultural Sciences; Shanghai Jiao Tong University","inventors":["焦杰然","张卫东","方献平","常丽英","孙志坚","胡智焕","贺世伟"],"publication_date":"2025-10-03","filing_date":"2025-08-20","priority_date":"2025-08-20","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G06F16/33295","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3344","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/3331","G06F16/334","G06F16/3346","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0499","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04","G06N5/041"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120744065A/en"},{"publication_number":"CN120744233A","title":"Collaborative filtering recommendation method for semantic enhancement of large language model based on contrast learning","abstract":"The invention discloses a collaborative filtering recommendation method for semantic enhancement of a large language model based on contrast learning, which belongs to the field of recommendation systems and comprises the steps of constructing a user-project bipartite graph, providing user-article semantic information collaboration information through GNN, extracting user-article semantic information by using prompt words, generating a semantic neighbor expansion view by using a deterministic topology view enhancement strategy, generating a semantic neighbor reconstruction view, aligning collaboration with semantic information, aligning a double-view structure and integrally training a loss function.","assignee":"Yanshan University","inventors":["宫继兵","谌君泽","赵祎","陈乐�","李林轩"],"publication_date":"2025-10-03","filing_date":"2025-06-24","priority_date":"2025-06-24","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G06F16/9535","G","G06","G06F","G06F16/00","G06F16/90","G06F16/95","G06F16/953","G06F16/9536","G","G06","G06F","G06F40/00","G06F40/30","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/042","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120744233A/en"},{"publication_number":"CN120752631A","title":"Condition-based asset management","abstract":"Methods, systems, and devices for predicting the success of a maintenance cycle performed on an asset based on a plurality of operating parameters. The predictive model may be trained and tested based on the plurality of operating parameters. The predictive model may be configured to output a prediction indicative of the degree of success of the maintenance cycle.","assignee":"Axlon Co","inventors":["P-C·陈","C·吉拉德","A·塔耶比","J·斯维尔泰克","G·莱恩豪泽尔","A·阿加沃"],"publication_date":"2025-10-03","filing_date":"2024-01-03","priority_date":"2023-01-04","cpc_codes":["G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0218","G05B23/0243","G05B23/0254","G","G05","G05B","G05B23/00","G05B23/02","G05B23/0205","G05B23/0259","G05B23/0283","G","G06","G06N","G06N20/00"],"country":"CN","kind":"application","source_url":"https://patents.google.com/patent/CN120752631A/en"},{"publication_number":"KR20250143716A","title":"Artificial intelligence-based meeting analysis and participant evaluation system","abstract":"본 발명은 다음과 같은 구성 요소를 포함한다. 1. 회의 기록 모듈: 회의실에 설치된 마이크와 카메라를 통해 회의의 음성 및 영상 데이터를 실시간으로 수집하고 저장하는 장치. 2. 데이터 전처리 모듈: 수집된 음성 데이터를 텍스트로 변환(STT: Speech-to-Text)하고, 영상 데이터에서 발언자의 얼굴 및 제스처 정보를 추출하는 모듈. 3. AI 분석 모듈: A. 자연어 처리(NLP) 엔진: 변환된 텍스트를 분석하여 발언의 내용, 감정, 논리적 흐름을 파악한다. 발언 내용에 따라 '혁신적 제안', '실행 가능성 분석', '협력적 태도', '비생산적 비판' 등의 태그를 자동으로 부여한다. B. 음성 분석 엔진: 발언자의 목소리 톤, 발화 속도, 감정 등을 분석하여 발언의 적극성 및 자신감을 평가한다. C. 비디오 분석 엔진: 발언자의 표정, 제스처, 시선 처리 등을 분석하여 회의 참여도와 몰입도를 측정한다. 4. 점수 산정 모듈: AI 분석 모듈에서 도출된 데이터를 바탕으로 각 참여자별 기여도 점수를 산정한다. 점수 산정은 미리 설정된 가중치(예: 혁신 기여도 40%, 실행 가능성 30%, 협력적 태도 20%, 비생산적 발언 -10%)에 따라 이루어진다. 5. 데이터베이스 및 평가 관리 모듈: A. 참여자별 회의 기여 점수, 프로젝트 성과 데이터 등 개인의 모든 성과 데이터를 통합하여 저장하고 관리한다. B. 저장된 데이터를 기반으로 분기별 또는 연간 단위로 개인의 성과를 자동으로 평가하여 리포트를 생성한다. C. 미리 설정된 기준에 따라 하위 성과자를 자동으로 식별하고, 상위 성과자에게 보상 계획을 제안하는 등의 인사 관리를 지원한다. 6. 사용자 인터페이스(UI): 참여자 개인이 자신의 기여도 점수와 분석 리포트를 확인할 수 있는 대시보드 및 관리자가 전체 조직의 성과 데이터를 한눈에 볼 수 있는 인터페이스를 제공한다. [색인어] 회의시스템, 인공지능 회의평가, 인공지능 인사평가, 생산적인 회의결론 도출 The present invention comprises the following components: 1. Meeting recording module: A device that collects and stores meeting audio and video data in real time through microphones and cameras installed in the conference room. 2. Data preprocessing module: A module that converts collected voice data into text (STT: Speech-to-Text) and extracts the speaker's face and gesture information from video data. 3. AI Analysis Module: A. Natural Language Processing (NLP) Engine: Analyzes the converted text to understand the content, sentiment, and logical flow of the speech. Depending on the content of the speech, tags such as \"innovative proposal,\"\"feasibilityanalysis,\"\"collaborativeattitude,\" and \"unproductive criticism\" are automatically assigned. B. Voice Analysis Engine: Analyzes the speaker's voice tone, speaking speed, emotion, etc. to evaluate the speaker's assertiveness and confidence in speaking. C. Video Analysis Engine: Measures meeting participation and immersion by analyzing the speaker's facial expressions, gestures, and eye contact. 4. Scoring Module: Based on data derived from the AI analysis module, a contribution score is calculated for each participant. Scoring is based on preset weights (e.g., 40% for innovation contribution, 30% for feasibility, 20% for collaborative attitude, -10% for unproductive comments). 5. Database and Assessment Management Module: A. All individual performance data, including participant contribution scores and project performance data, is integrated, stored, and managed. B. Automatically evaluate individual performance on a quarterly or annual basis based on stored data and generate reports. C. Supports human resource management, such as automatically identifying low performers based on preset criteria and suggesting compensation plans to high performers. 6. User Interface (UI): Provides a dashboard where individual participants can check their contribution scores and analysis reports, and an interface where managers can view the performance data of the entire organization at a glance. [index word] Meeting systems, AI meeting evaluation, AI personnel evaluation, and productive meeting conclusions.","assignee":"김동환","inventors":["김동환"],"publication_date":"2025-10-02","filing_date":"2025-09-15","priority_date":"2025-09-15","cpc_codes":["G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06393","G","G06","G06F","G06F40/00","G06F40/20","G","G06","G06N","G06N20/00","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/063","G06Q10/0639","G06Q10/06398","G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/105","G","G10","G10L","G10L15/00","G10L15/26"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250143716A/en"},{"publication_number":"AU2025230826A1","title":"Transcriptome Deconvolution Of Metastatic Tissue Samples","abstract":"A platform for transcriptome deconvolution of gene expression data is provided and may be used in assessing metastatic cancer samples. The deconvolution is performed using an unsupervised clustering technique, such as grade of membership, that allows for samples to be assigned to multiple clusters during a training process. A deconvolution gene expression model is generated as a result and is used for accurate assess of metastases in subsequent samples. (FIG. 1)","assignee":"Tempus AI Inc","inventors":["Mathew BARBER","Catherine Igartua","Kaanan Shah"],"publication_date":"2025-10-02","filing_date":"2025-09-15","priority_date":"2018-12-31","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/12","G06N3/123","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0895","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G16","G16B","G16B25/00","G16B25/10","G","G16","G16B","G16B30/00","G16B30/10","G","G16","G16B","G16B40/00","G16B40/20","G","G16","G16B","G16B40/00","G16B40/30","G","G16","G16H","G16H10/00","G16H10/40","G","G16","G16H","G16H50/00","G16H50/20","G","G06","G06N","G06N20/00","G06N20/10","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N5/00","G06N5/01","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G06N5/025","G","G06","G06N","G06N7/00","G06N7/01"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025230826A1/en"},{"publication_number":"AU2025230736A1","title":"Systems and Methods of Generating Medical Concordance Scores","abstract":"Examples may provide an electronic neural network that has been trained on a set of training data that comprises a plurality of reference subject medical data sets that are each labeled with a medical determination and are each assigned a ground truth concordance score generated by a plurality of experts in which a value of a given ground truth concordance score comprises a fraction of the plurality of experts, if any, that are in accord with the medical determination label of a given reference subject medical data set in the plurality of reference subject medical data sets. The electronic neural network is configured to provide an output concordance score of the medical determination being indicated by a test subject medical data set.","assignee":"Proscia Inc","inventors":["Sean GRULLON","Julianna IANNI","Vaughn SPURRIER"],"publication_date":"2025-10-02","filing_date":"2025-09-11","priority_date":"2022-02-04","cpc_codes":["G","G16","G16H","G16H50/00","G16H50/20","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G16","G16H","G16H10/00","G16H10/60","G","G16","G16H","G16H15/00","G","G16","G16H","G16H20/00","G16H20/10","G","G16","G16H","G16H30/00","G16H30/40","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/70"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025230736A1/en"},{"publication_number":"AU2025230677A1","title":"System for counting number of game tokens","abstract":"MARKED-UP COPY MARKED-UP COPY MARKED-UP COPY MARKED-UP COPY MARKED-UP COPY Provided is a chip recognition system having a mechanism for recognizing the colors of chips from an image and for enabling counting of the number of the chips even under different illumination environments. A chip (W) is configured to at least partially have a specific color (121) indicating the value of the chip (W). A chip recognition system (10) comprises: a recording apparatus (11) that records, as an image, the state of the chip W by using a camera (212); an image analysis apparatus (14) that, by analyzing the recorded image, recognizes at least two colors including the specific color (121) and a reference color (4a) existing in the image but being different from the specific color (121); and a recognition apparatus (12) including at least an artificial intelligence device (12a) that specifies the specific color (121) of the chip (W) by using the image analysis result by the image analysis apparatus (14), wherein a plurality of images of the chip (W) and the reference color (121) illuminated at different brightnesses are taught, as teacher data, to the artificial intelligence device (12a) of the recognition apparatus (12).","assignee":"Angel Group Co Ltd","inventors":["Yasushi Shigeta"],"publication_date":"2025-10-02","filing_date":"2025-09-10","priority_date":"2017-02-21","cpc_codes":["G","G06","G06V","G06V10/00","G06V10/70","G06V10/77","G06V10/778","G06V10/7784","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06M","G06M11/00","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06Q","G06Q50/00","G06Q50/34","G","G06","G06T","G06T7/00","G06T7/0002","G","G06","G06T","G06T7/00","G06T7/70","G06T7/73","G","G06","G06T","G06T7/00","G06T7/90","G","G06","G06V","G06V10/00","G06V10/10","G06V10/12","G06V10/14","G06V10/141","G","G06","G06V","G06V10/00","G06V10/20","G06V10/22","G","G06","G06V","G06V10/00","G06V10/20","G06V10/22","G06V10/225","G","G06","G06V","G06V10/00","G06V10/20","G06V10/24","G06V10/245","G","G06","G06V","G06V10/00","G06V10/40","G06V10/56","G","G06","G06V","G06V20/00","G06V20/60","G06V20/64","G","G07","G07D","G07D7/00","G07D7/20","G07D7/2016","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3202","G07F17/3216","G07F17/322","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3241","G","G07","G07F","G07F17/00","G07F17/32","G07F17/3244","G07F17/3248","H","H04","H04N","H04N5/00","H04N5/76","H04N5/765","H04N5/77","G","G06","G06T","G06T2207/00","G06T2207/10","G06T2207/10024","G","G06","G06T","G06T2207/00","G06T2207/20","G06T2207/20081"],"country":"AU","kind":"application","source_url":"https://patents.google.com/patent/AU2025230677A1/en"},{"publication_number":"US20250307604A1","title":"Adaptively training of neural networks via an intelligent learning management system","abstract":"Computer systems and computer-implemented methods train a neural network, by:(a) computing for each datum in a set of training data, activation values for nodes in the neural network and estimates of partial derivatives of an objective function for the neural network for the nodes in the neural network; (b) selecting a target node of the neural network and/or a target datum in the set of training data; (c) selecting a target-specific improvement model for the neural network, wherein the target-specific improvement model, when added to the neural network, improves performance of the neural network for the target node and/or the target datum, as the case may be; (d) training the target-specific improvement model; (e) merging the target-specific improvement model with the neural network to form an expanded neural network; and (f) training the expanded neural network.","assignee":"D5AI LLC","inventors":["James K. Baker","Bradley J. Baker"],"publication_date":"2025-10-02","filing_date":"2025-06-12","priority_date":"2020-08-20","cpc_codes":["G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/048","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/082","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/09","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/0985","G","G06","G06N","G06N5/00","G06N5/01"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250307604A1/en"},{"publication_number":"US20250307716A1","title":"Systems and methods for federated learning optimization via cluster feedback","abstract":"A method for generating a cluster-based machine learning model based on federated learning with cluster feedback includes providing a current machine learning model to a plurality of user devices that train the current machine learning model, receiving respective model states, generating updated model states, causing the plurality of user devices to obtain a respective instance of an updated machine learning model based on the updated model states, receiving an applicability feedback for the updated machine learning model for each of the plurality of user devices, determining a plurality of user clusters including a subset of the plurality of user devices, identifying a first user cluster and a second user cluster, the first user cluster having a higher cluster applicability feedback than the second user cluster, receiving the additional model states from the clusters and updating the updated machine learning model to generate the cluster-based machine learning model.","assignee":"Capital One Services LLC","inventors":["Jeremy Goodsitt","Christopher Wallace","Grant Eden","Anh Truong","Austin Walters"],"publication_date":"2025-10-02","filing_date":"2025-06-16","priority_date":"2021-12-07","cpc_codes":["G","G06","G06N","G06N20/00","G06N20/20","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06N","G06N7/00","G06N7/01","G","G06","G06N","G06N20/00"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250307716A1/en"},{"publication_number":"US20250307321A1","title":"Causal reasoning system","abstract":"A research assistant system described herein includes a research assistant tool and associated components and a graphical user interface to guide user input to research, discover, and evidence answers for complex research questions. The research assistant system may include the graphical user interface (“GUI” or “user interface”) for presentation on a user device associated with a user. The user interface may provide prompts and guidance for collaboration and exploration of research concepts iteratively. A concept may include a search term, entities, and/or propositions/statements.","assignee":"Bridgewater Associates Ec Ip LLC","inventors":["David A. Ferrucci","Aditya A. Kalyanpur","Victor BARRES","Clifton James McFate","Jose Barrera","Kailash Karthik Saravanakumar","Maksim Eremeev","Jennifer Chu-Carroll"],"publication_date":"2025-10-02","filing_date":"2025-03-27","priority_date":"2024-03-28","cpc_codes":["G","G06","G06F","G06F16/00","G06F16/90","G06F16/93","G","G06","G06F","G06F16/00","G06F16/30","G06F16/33","G06F16/332","G06F16/3329","G","G06","G06F","G06F9/00","G06F9/06","G06F9/44","G06F9/451","G","G06","G06N","G06N5/00","G06N5/02","G06N5/022","G","G06","G06N","G06N5/00","G06N5/04"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250307321A1/en"},{"publication_number":"US20250306554A1","title":"AI-Based Energy Edge Platform, Systems, and Methods Having a Digital Twin of a Mining Environment","abstract":"An AI-based platform for enabling intelligent orchestration and management of power and energy is disclosed. The platform includes a digital twin system having a digital twin of a mining environment. The digital twin includes at least one parameter that is detected by a sensor of the mining environment. In some disclosed embodiments, the at least one parameter is associated with one or more of an unmined portion of the mining environment a mining of materials from the mining environment, a smart container event involving a smart container associated with the mining environment, a physiological status of a miner associated with the mining environment, a transaction-related event associated with the mining environment, and a compliance of the mining environment with one or more contractual, regulatory, and/or legal policies.","assignee":"Strong Force Ee Portfolio 2022 LLC","inventors":["Charles Howard Cella","Andrew Cardno"],"publication_date":"2025-10-02","filing_date":"2025-03-24","priority_date":"2021-11-23","cpc_codes":["G","G06","G06Q","G06Q50/00","G06Q50/06","G","G01","G01R","G01R21/00","G01R21/133","G","G05","G05B","G05B13/00","G05B13/02","G05B13/0265","G","G05","G05B","G05B13/00","G05B13/02","G05B13/04","G","G05","G05B","G05B13/00","G05B13/02","G05B13/04","G05B13/042","G","G05","G05B","G05B19/00","G05B19/02","G05B19/04","G05B19/042","G","G06","G06F","G06F1/00","G06F1/26","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/213","G","G06","G06F","G06F18/00","G06F18/20","G06F18/21","G06F18/214","G","G06","G06F","G06F18/00","G06F18/20","G06F18/24","G06F18/245","G06F18/2453","G","G06","G06N","G06N10/00","G","G06","G06N","G06N10/00","G06N10/80","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N5/00","G06N5/04","G06N5/043","G","G06","G06Q","G06Q10/00","G06Q10/04","G","G06","G06Q","G06Q10/00","G06Q10/06","G06Q10/067","G","G06","G06Q","G06Q30/00","G06Q30/018","G","G06","G06Q","G06Q50/00","G06Q50/02","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/26","G","G06","G06Q","G06Q99/00","G","G06","G06V","G06V10/00"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250306554A1/en"},{"publication_number":"US20250308083A1","title":"Reference image structure match using diffusion models","abstract":"A method, apparatus, non-transitory computer readable medium, and system for image processing include obtaining a structural input indicating a target spatial structure, encoding, using a condition encoder, the structural input to obtain a structural encoding representing the target spatial structure, and generating, using an image generation model, a synthetic image based on the structural encoding, where the synthetic image depicts an object having the target spatial structure.","assignee":"Adobe Inc","inventors":["Sachin Madhav Kelkar","Fengbin Chen","Hareesh Ravi","Zhifei Zhang","Ajinkya Gorakhnath Kale","Zhe Lin"],"publication_date":"2025-10-02","filing_date":"2024-11-14","priority_date":"2024-03-26","cpc_codes":["G","G06","G06T","G06T11/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0475","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08"],"country":"US","kind":"application","source_url":"https://patents.google.com/patent/US20250308083A1/en"},{"publication_number":"KR20250143294A","title":"Ferroelectric nanoparticle capacitor for non-binary logics and method of operation","abstract":"강유전체 나노입자 커패시터 디바이스는 서로 전기적으로 절연된 전도성 요소들의 쌍, 및 쌍의 전도성 요소들 사이에 배열된 강유전체 나노입자들을 포함한다. 강유전체 나노입자들은 상이한 총 강유전체 분극들을 갖는 적어도 3개의 분극 상태를 제공하도록 적응된다. A ferroelectric nanoparticle capacitor device comprises a pair of electrically insulated conductive elements, and ferroelectric nanoparticles arranged between the pair of conductive elements. The ferroelectric nanoparticles are adapted to provide at least three polarization states having different total ferroelectric polarizations.","assignee":"테라 퀀텀 아게","inventors":["안나 라줌나야","유리 티코노브","이고르 루키안척","발레리 비노쿠르"],"publication_date":"2025-10-01","filing_date":"2025-09-22","priority_date":"2022-10-11","cpc_codes":["H","H01","H01G","H01G7/00","H01G7/06","H","H01","H01G","H01G4/00","H01G4/40","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/06","G06N3/063","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G11","G11C","G11C11/00","G11C11/21","G11C11/22","G11C11/221","G","G11","G11C","G11C11/00","G11C11/21","G11C11/22","G11C11/225","G11C11/2275","G","G11","G11C","G11C11/00","G11C11/54","G","G11","G11C","G11C11/00","G11C11/56","G11C11/5657","H","H01","H01G","H01G4/00","H01G4/002","H01G4/018","H01G4/06","H01G4/08","H01G4/10","H","H01","H01G","H01G4/00","H01G4/33","H","H03","H03K","H03K19/00","H03K19/02","H03K19/185","H","H10","H10B","H10B51/00","H10B51/30","H","H10","H10B","H10B53/00","H10B53/30","H","H10","H10D","H10D1/00","H10D1/01","H10D1/041","H","H10","H10D","H10D1/00","H10D1/60","H10D1/68","H","H10","H10D","H10D1/00","H10D1/60","H10D1/68","H10D1/682","H","H10","H10D","H10D1/00","H10D1/60","H10D1/68","H10D1/692","H","H10","H10D","H10D1/00","H10D1/60","H10D1/68","H10D1/692","H10D1/694","H","H10","H10D","H10D62/00","H10D62/10","H10D62/117","H10D62/118","H","H10"],"country":"KR","kind":"application","source_url":"https://patents.google.com/patent/KR20250143294A/en"},{"publication_number":"MX2025011026A","title":"Communication methods and apparatuses, device, chip and storage medium","abstract":"Provided in the embodiments of the present application are communication methods. A method comprises: receiving first information from a first network element, the first information being used for indicating at least one second node that satisfies a condition for executing a VFL task; and determining from the at least one second node at least one participant node that participates in the VFL task. In the method, a first node that initiates a VFL task can receive first information from a first network element, and can know, by means of the first information, at least one second node that satisfies a condition for executing the VFL task, so as to ensure that the first node can find suitable nodes for performing the VFL task.","assignee":"Guangdong Oppo Mobile Telecommunications Corp Ltd","inventors":["Jingran Chen"],"publication_date":"2025-10-01","filing_date":"2025-09-18","priority_date":"2023-03-31","cpc_codes":["H","H04","H04L","H04L41/00","H04L41/16","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/098","H","H04","H04L","H04L41/00","H04L41/08","H04L41/0803","H04L41/0806","H","H04","H04L","H04L41/00","H04L41/08","H04L41/085","H04L41/0853","H","H04","H04W","H04W60/00","H04W60/04","H","H04","H04W","H04W8/00","H04W8/22","H","H04","H04W","H04W8/00","H04W8/22","H04W8/24","H","H04","H04L","H04L41/00","H04L41/08","H04L41/085","H","H04","H04L","H04L41/00","H04L41/14","H04L41/145"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2025011026A/en"},{"publication_number":"MX2025010931A","title":"Biomarker compositions and methods of use thereof","abstract":"The present disclosure provides methods and kits for identifying and treating individuals at risk of or suffering from amyloid transthyretin cardiomyopathy. In general, detection or measurement of one or more biomarkers, such as TnI, PKM1, PKM2, NT-proBNP, RBP4, DCN, TIMP2, SMOC-2, NfL, or combinations thereof, assists in the identification of amyloid transthyretin cardiomyopathy. The present disclosure also provides methods for selecting patients for treatment of amyloid transthyretin cardiomyopathy, such as with transthyretin stabilizing agents.","assignee":"Siemens Healthcare Diagnostics Inc","inventors":["Chris Green","Mark Baumeister","Arejas James Uzgiris"],"publication_date":"2025-10-01","filing_date":"2025-09-15","priority_date":"2023-03-15","cpc_codes":["G","G16","G16H","G16H10/00","G16H10/40","G","G01","G01N","G01N33/00","G01N33/48","G01N33/50","G01N33/68","G01N33/6893","G","G06","G06N","G06N7/00","G06N7/01","G","G16","G16H","G16H30/00","G16H30/40","G","G16","G16H","G16H40/00","G16H40/60","G16H40/67","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/30","G","G16","G16H","G16H50/00","G16H50/70","G","G01","G01N","G01N2800/00","G01N2800/32","G01N2800/325","G","G01","G01N","G01N2800/00","G01N2800/32","G01N2800/326","G","G01","G01N","G01N2800/00","G01N2800/60","G","G06","G06N","G06N20/00"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2025010931A/en"},{"publication_number":"MX2025010760A","title":"Methods of developing cancer diagnostic models and uses thereof in developing cancer detection methods","abstract":"A cancer diagnostic model development method is provided, which includes constructing a training set and building a diagnostic model thereon. The training set comprises miRNA expression profiles from non-cancer subjects and cancer patients with two or more cancer types, and building the diagnostic model includes calculating a diagnostic index based on a selected miRNA biomarker set obtained according to the rankings of miRNAs in a differential expression analysis of the miRNA expression profiles in the training set. Methods for detecting a cancer of interest by means of such developed diagnostic model are also provided. A 4-miRNA based diagnostic model demonstrates high performances in a validation set, capable of achieving sensitivity of â¿¥ 0.98 while maintaining specificity of 0.99 in detecting multiple cancers including lung cancer, gastric cancer, biliary tract cancer, bladder cancer, prostate cancer, and glioma.","assignee":"Mironcol Diagnostics Ltd","inventors":["Jason Zhang","Hai Hu"],"publication_date":"2025-10-01","filing_date":"2025-09-11","priority_date":"2023-03-13","cpc_codes":["G","G16","G16B","G16B40/00","G16B40/20","C","C12","C12Q","C12Q1/00","C12Q1/68","C12Q1/6876","C12Q1/6883","C12Q1/6886","G","G06","G06N","G06N20/00","G","G16","G16B","G16B25/00","G16B25/10","G","G16","G16H","G16H50/00","G16H50/20","C","C12","C12Q","C12Q2600/00","C12Q2600/158","C","C12","C12Q","C12Q2600/00","C12Q2600/178"],"country":"MX","kind":"application","source_url":"https://patents.google.com/patent/MX2025010760A/en"},{"publication_number":"EP4623821A2","title":"System and method for decision support using lifestyle factors","abstract":"Systems and methods are provided relating to open loop decision-making for management of diabetes. People with diabetes face many problems in controlling their glucose because of the complex interactions between food, insulin, exercise, stress, activity, and other physiological and environmental conditions. Established principles of management of glucose sometimes are not adequate because there is a significant amount of variability in how different conditions impact different individuals and what actions might be effective for them. Accordingly, systems and methods according to present principles minimize the impact of the vagaries of diabetes on individuals, i.e., by looking for patterns and tendencies of an individual and customizing the management to that individual. Consequently, the same reduces the uncertainty that diabetes typically is associated with and improves quality of life.","assignee":"Dexcom Inc","inventors":["Anna Leigh Davis","Jr. Esteban Cabrera","Nathaniel David Heintzman","Andrew Attila Pal","Naresh C. Bhavaraju","Alexandra Elena CONSTANTIN","Lauren Hruby Jepson","Eli Reihman","Jennifer Blackwell","Basab Dattaray","Apurv Ullas Kamath","Tomas C. WALKER","Leif N. Bowman","Rian DRAEGER","Katherine Yerre Koehler"],"publication_date":"2025-10-01","filing_date":"2017-01-26","priority_date":"2016-02-01","cpc_codes":["A","A61","A61M","A61M5/00","A61M5/14","A61M5/142","A61M5/14244","A","A61","A61B","A61B5/00","A61B5/0002","A61B5/0015","A","A61","A61B","A61B5/00","A61B5/145","A61B5/14503","A","A61","A61B","A61B5/00","A61B5/145","A61B5/14532","A","A61","A61B","A61B5/00","A61B5/48","A61B5/4836","A61B5/4839","A","A61","A61B","A61B5/00","A61B5/72","A61B5/7235","A61B5/7264","A","A61","A61M","A61M5/00","A61M5/14","A61M5/142","G","G06","G06N","G06N20/00","G","G06","G06N","G06N5/00","G06N5/04","G06N5/048","G","G16","G16H","G16H20/00","G16H20/10","G16H20/17","G","G16","G16H","G16H20/00","G16H20/60","G","G16","G16H","G16H40/00","G16H40/60","G16H40/63","G","G16","G16H","G16H50/00","G16H50/20","G","G16","G16H","G16H50/00","G16H50/70","A","A61","A61M","A61M5/00","A61M5/14","A61M5/142","A61M2005/14208","A","A61","A61M","A61M2205/00","A61M2205/50","A61M2205/502","A","A61","A61M","A61M2205/00","A61M2205/58","A61M2205/581","A","A61","A61M","A61M5/00","A61M5/178","A61M5/20","G","G06","G06Q","G06Q10/00","G06Q10/10","G06Q10/109","G","G16","G16H","G16H20/00","G16H20/30","G","G16"],"country":"EP","kind":"application","source_url":"https://patents.google.com/patent/EP4623821A2/en"},{"publication_number":"TWI899647B","title":"Method, module, apparatus, and system for compute optimizations for low precision machine learning operations, and computer-readable medium","abstract":"One embodiment provides a general-purpose graphics processing unit comprising a dynamic precision floating-point unit including a control unit having precision tracking hardware logic to track an available number of bits of precision for computed data relative to a target precision, wherein the dynamic precision floating-point unit includes computational logic to output data at multiple precisions.","assignee":"美商英特爾股份有限公司","inventors":["亞奇雪克 亞布","班 亞西鮑彿","金德煥","艾蒙斯特阿法 歐德亞麥德維爾","屏 唐","陳曉明","巴拉斯 拉克斯曼","凱文 尼爾斯","麥可 史崔克蘭","歐塔 寇克","麥克 麥佛森","喬伊迪普 雷","姚安邦","馬立偉","琳達 赫德","莎拉 班索艾","約翰 威斯特"],"publication_date":"2025-10-01","filing_date":"2018-02-22","priority_date":"2017-04-28","cpc_codes":["G","G06","G06T","G06T1/00","G06T1/20","G","G06","G06F","G06F12/00","G06F12/02","G06F12/08","G06F12/0802","G06F12/0806","G06F12/0811","G","G06","G06F","G06F15/00","G06F15/16","G06F15/163","G06F15/167","G","G06","G06F","G06F15/00","G06F15/16","G06F15/163","G06F15/17","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/483","G","G06","G06F","G06F7/00","G06F7/38","G06F7/48","G06F7/57","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/30003","G06F9/30007","G06F9/3001","G06F9/30014","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/30181","G06F9/30185","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/38","G06F9/3861","G06F9/3863","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/38","G06F9/3867","G","G06","G06F","G06F9/00","G06F9/06","G06F9/30","G06F9/38","G06F9/3885","G06F9/3887","G","G06","G06F","G06F9/00","G06F9/06","G06F9/46","G06F9/50","G06F9/5005","G06F9/5027","G06F9/5044","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02"],"country":"TW","kind":"application","source_url":"https://patents.google.com/patent/TWI899647B/en"},{"publication_number":"EP4625303A1","title":"Data processing method and related device","abstract":"A data processing method is applied to recommendation of a learning order of knowledge points. The method includes: obtaining feature representations of a plurality of knowledge points and a first learning state of a user, where the first learning state indicates a user's level of mastery of a learned knowledge point, and the plurality of knowledge points are different from the learned knowledge point; and obtaining, by using a decoder based on the feature representations of the plurality of knowledge points, the first learning state, and a learning objective, a learning order corresponding to the plurality of knowledge points, where the decoder is configured to identify a relationship between the feature representations of the plurality of knowledge points and the learning objective in the first learning state. In this application, when the learning order of the knowledge points is determined, the learning order of the plurality of knowledge points is determined in a sequence generation manner based on a current learning state of the user, and a sequence is not necessarily selected from a large quantity of learning orders. In this way, a finally determined learning order is not limited to a path used by another user for learning in the past, thereby improving a personalization degree of a learning path and learning effect.","assignee":"Huawei Technologies Co Ltd","inventors":["Wei Xia","Xianyu Chen","Weiwen Liu","Ruiming TANG","Weinan Zhang","Yong Yu"],"publication_date":"2025-10-01","filing_date":"2023-12-08","priority_date":"2022-12-09","cpc_codes":["G","G09","G09B","G09B5/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/044","G06N3/0442","G","G06","G06F","G06F18/00","G06F18/20","G06F18/25","G06F18/253","G","G06","G06N","G06N20/00","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/045","G06N3/0455","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/092","G","G09","G09B","G09B7/00","G09B7/02","G09B7/04","G","G09","G09B","G09B7/00","G09B7/06","G09B7/08","G","G06","G06N","G06N3/00","G06N3/02","G06N3/04","G06N3/0464","G","G06","G06N","G06N3/00","G06N3/02","G06N3/08","G06N3/084","G","G06","G06Q","G06Q50/00","G06Q50/10","G06Q50/20","G06Q50/205"],"country":"EP","kind":"application","source_url":"https://patents.google.com/patent/EP4625303A1/en"}]}